We introduce a light steering technology that operates at megahertz frequencies, has no moving parts, and costs less than a hundred dollars. Our technology can benefit many projector and imaging systems that critically rely on high-speed, reliable, low-cost, and wavelength-independent light steering, including laser scanning projectors, LiDAR sensors, and fluorescence microscopes. Our technology uses ultrasound waves to generate a spatiotemporally-varying refractive index field inside a compressible medium, such as water, turning the medium into a dynamic traveling lens. By controlling the electrical input of the ultrasound transducers that generate the waves, we can change the lens, and thus steer light, at the speed of sound (1.5 km/s in water). We build a physical prototype of this technology, use it to realize different scanning techniques at megahertz rates (three orders of magnitude faster than commercial alternatives such as galvo mirror scanners), and demonstrate proof-of-concept projector and LiDAR applications. To encourage further innovation towards this new technology, we derive the theory for its fundamental limits and develop a physically-accurate simulator for virtual design. Our technology offers a promising solution for achieving high-speed and low-cost light steering in a variety of applications.

Dynamic radiance field reconstruction methods aim to model the time-varying structure and appearance of a dynamic scene. Existing methods, however, assume that accurate camera poses can be reliably estimated by Structure from Motion (SfM) algorithms. These methods, thus, are unreliable as SfM algorithms often fail or produce erroneous poses on challenging videos with highly dynamic objects, poorly textured surfaces, and rotating camera motion. We address this issue by jointly estimating the static and dynamic radiance fields along with the camera parameters (poses and focal length). We demonstrate the robustness of our approach via extensive quantitative and qualitative experiments. Our results show favorable performance over the state-of-the-art dynamic view synthesis methods.

Recent works such as BARF and GARF can bundle adjust camera poses with neural radiance fields (NeRF) which is based on coordinate-MLPs. Despite the impressive results, these methods cannot be applied to Generalizable NeRFs (GeNeRFs) which require image feature extractions that are often based on more complicated 3D CNN or transformer architectures. In this work, we first analyze the difficulties of jointly optimizing camera poses with GeNeRFs, and then further propose our DBARF to tackle these issues. Our DBARF which bundle adjusts camera poses by taking a cost feature map as an implicit cost function can be jointly trained with GeNeRFs in a self-supervised manner. Unlike BARF and its follow-up works, which can only be applied to per-scene optimized NeRFs and need accurate initial camera poses with the exception of forward-facing scenes, our method can generalize across scenes and does not require any good initialization. Experiments show the effectiveness and generalization ability of our DBARF when evaluated on real-world datasets. Our code is available at https://aibluefisher.github.io/dbarf.

We propose VDN-NeRF, a method to train neural radiance fields (NeRFs) for better geometry under non-Lambertian surface and dynamic lighting conditions that cause significant variation in the radiance of a point when viewed from different angles. Instead of explicitly modeling the underlying factors that result in the view-dependent phenomenon, which could be complex yet not inclusive, we develop a simple and effective technique that normalizes the view-dependence by distilling invariant information already encoded in the learned NeRFs. We then jointly train NeRFs for view synthesis with view-dependence normalization to attain quality geometry. Our experiments show that even though shape-radiance ambiguity is inevitable, the proposed normalization can minimize its effect on geometry, which essentially aligns the optimal capacity needed for explaining view-dependent variations. Our method applies to various baselines and significantly improves geometry without changing the volume rendering pipeline, even if the data is captured under a moving light source. Code is available at: https://github.com/BoifZ/VDN-NeRF.

Neural Radiance Fields (NeRFs) are a powerful representation for modeling a 3D scene as a continuous function. Though NeRF is able to render complex 3D scenes with view-dependent effects, few efforts have been devoted to exploring its limits in a high-resolution setting. Specifically, existing NeRF-based methods face several limitations when reconstructing high-resolution real scenes, including a very large number of parameters, misaligned input data, and overly smooth details. In this work, we conduct the first pilot study on training NeRF with high-resolution data and propose the corresponding solutions: 1) marrying the multilayer perceptron (MLP) with convolutional layers which can encode more neighborhood information while reducing the total number of parameters; 2) a novel training strategy to address misalignment caused by moving objects or small camera calibration errors; and 3) a high-frequency aware loss. Our approach is nearly free without introducing obvious training/testing costs, while experiments on different datasets demonstrate that it can recover more high-frequency details compared with the current state-of-the-art NeRF models. Project page: https://yifanjiang19.github.io/alignerf.

Research on neural radiance fields (NeRFs) for novel view generation is exploding with new models and extensions. However, a question that remains unanswered is what happens in underwater or foggy scenes where the medium strongly influences the appearance of objects. Thus far, NeRF and its variants have ignored these cases. However, since the NeRF framework is based on volumetric rendering, it has inherent capability to account for the medium’s effects, once modeled appropriately. We develop a new rendering model for NeRFs in scattering media, which is based on the SeaThru image formation model, and suggest a suitable architecture for learning both scene information and medium parameters. We demonstrate the strength of our method using simulated and real-world scenes, correctly rendering novel photorealistic views underwater. Even more excitingly, we can render clear views of these scenes, removing the medium between the camera and the scene and reconstructing the appearance and depth of far objects, which are severely occluded by the medium. Our code and unique datasets are available on the project’s website.

Neural Radiance Fields (NeRF) have attracted significant attention due to their ability to synthesize novel scene views with great accuracy. However, inherent to their underlying formulation, the sampling of points along a ray with zero width may result in ambiguous representations that lead to further rendering artifacts such as aliasing in the final scene. To address this issue, the recent variant mip-NeRF proposes an Integrated Positional Encoding (IPE) based on a conical view frustum. Although this is expressed with an integral formulation, mip-NeRF instead approximates this integral as the expected value of a multivariate Gaussian distribution. This approximation is reliable for short frustums but degrades with highly elongated regions, which arises when dealing with distant scene objects under a larger depth of field. In this paper, we explore the use of an exact approach for calculating the IPE by using a pyramid-based integral formulation instead of an approximated conical-based one. We denote this formulation as Exact-NeRF and contribute the first approach to offer a precise analytical solution to the IPE within the NeRF domain. Our exploratory work illustrates that such an exact formulation (Exact-NeRF) matches the accuracy of mip-NeRF and furthermore provides a natural extension to more challenging scenarios without …

The success of the Neural Radiance Fields (NeRFs) for modeling and free-view rendering static objects has inspired numerous attempts on dynamic scenes. Current techniques that utilize neural rendering for facilitating free-view videos (FVVs) are restricted to either offline rendering or are capable of processing only brief sequences with minimal motion. In this paper, we present a novel technique, Residual Radiance Field or ReRF, as a highly compact neural representation to achieve real-time FVV rendering on long-duration dynamic scenes. ReRF explicitly models the residual information between adjacent timestamps in the spatial-temporal feature space, with a global coordinate-based tiny MLP as the feature decoder. Specifically, ReRF employs a compact motion grid along with a residual feature grid to exploit inter-frame feature similarities. We show such a strategy can handle large motions without sacrificing quality. We further present a sequential training scheme to maintain the smoothness and the sparsity of the motion/residual grids. Based on ReRF, we design a special FVV codec that achieves three orders of magnitudes compression rate and provides a companion ReRF player to support online streaming of long-duration FVVs of dynamic scenes. Extensive experiments demonstrate the effectiveness of ReRF for compactly representing dynamic radiance fields, enabling an unprecedented free-viewpoint …

In this paper, we present a new representation for neural radiance fields that accelerates both the training and the inference processes with VDB, a hierarchical data structure for sparse volumes. VDB takes both the advantages of sparse and dense volumes for compact data representation and efficient data access, being a promising data structure for NeRF data interpolation and ray marching. Our method, Plenoptic VDB (PlenVDB), directly learns the VDB data structure from a set of posed images by means of a novel training strategy and then uses it for real-time rendering. Experimental results demonstrate the effectiveness and the efficiency of our method over previous arts: First, it converges faster in the training process. Second, it delivers a more compact data format for NeRF data presentation. Finally, it renders more efficiently on commodity graphics hardware. Our mobile PlenVDB demo achieves 30+ FPS, 1280x720 resolution on an iPhone12 mobile phone. Check plenvdb.github.io for details.

We propose LIRF (Local Implicit Ray Function), a generalizable neural rendering approach for novel view rendering. Current generalizable neural radiance fields (NeRF) methods sample a scene with a single ray per pixel and may therefore render blurred or aliased views when the input views and rendered views observe scene content at different resolutions. To solve this problem, we propose LIRF to aggregate the information from conical frustums to construct a ray. Given 3D positions within conical frustums, LIRF takes 3D coordinates and the features of conical frustums as inputs and predicts a local volumetric radiance field. Since the coordinates are continuous, LIRF renders high-quality novel views at a continuously-valued scale via volume rendering. Besides, we predict the visible weights for each input view via transformer-based feature matching to improve the performance in occluded areas. Experimental results on real-world scenes validate that our method outperforms state-of-the-art methods on novel view rendering of unseen scenes at arbitrary scales.

Online reconstructing and rendering of large-scale indoor scenes is a long-standing challenge. SLAM-based methods can reconstruct 3D scene geometry progressively in real time but can not render photorealistic results. While NeRF-based methods produce promising novel view synthesis results, their long offline optimization time and lack of geometric constraints pose challenges to efficiently handling online input. Inspired by the complementary advantages of classical 3D reconstruction and NeRF, we thus investigate marrying explicit geometric representation with NeRF rendering to achieve efficient online reconstruction and high-quality rendering. We introduce SurfelNeRF, a variant of neural radiance field which employs a flexible and scalable neural surfel representation to store geometric attributes and extracted appearance features from input images. We further extend conventional surfel-based fusion scheme to progressively integrate incoming input frames into the reconstructed global neural scene representation. In addition, we propose a highly-efficient differentiable rasterization scheme for rendering neural surfel radiance fields, which helps SurfelNeRF achieve 10× speedups in training and inference time, respectively. Experimental results show that our method achieves the state-of-the-art 23.82 PSNR and 29.58 PSNR on ScanNet in feedforward inference and per-scene optimization settings, respectively.

We develop an effective point cloud rendering pipeline for novel view synthesis, which enables high fidelity local detail reconstruction, real-time rendering and user-friendly editing. In the heart of our pipeline is an adaptive frequency modulation module called Adaptive Frequency Net (AFNet), which utilizes a hypernetwork to learn the local texture frequency encoding that is consecutively injected into adaptive frequency activation layers to modulate the implicit radiance signal. This mechanism improves the frequency expressive ability of the network with richer frequency basis support, only at a small computational budget. To further boost performance, a preprocessing module is also proposed for point cloud geometry optimization via point opacity estimation. In contrast to implicit rendering, our pipeline supports high-fidelity interactive editing based on point cloud manipulation. Extensive experimental results on NeRF-Synthetic, ScanNet, DTU and Tanks and Temples datasets demonstrate the superior performances achieved by our method in terms of PSNR, SSIM and LPIPS, in comparison to the state-of-the-art.

Modeling and re-rendering dynamic 3D scenes is a challenging task in 3D vision. Prior approaches build on NeRF and rely on implicit representations. This is slow since it requires many MLP evaluations, constraining real-world applications. We show that dynamic 3D scenes can be explicitly represented by six planes of learned features, leading to an elegant solution we call HexPlane. A HexPlane computes features for points in spacetime by fusing vectors extracted from each plane, which is highly efficient. Pairing a HexPlane with a tiny MLP to regress output colors and training via volume rendering gives impressive results for novel view synthesis on dynamic scenes, matching the image quality of prior work but reducing training time by more than 100×. Extensive ablations confirm our HexPlane design and show that it is robust to different feature fusion mechanisms, coordinate systems, and decoding mechanisms. HexPlane is a simple and effective solution for representing 4D volumes, and we hope they can broadly contribute to modeling spacetime for dynamic 3D scenes.

We show how shadows can be efficiently generated in differentiable rendering of triangle meshes. Our central observation is that pre-filtered shadow mapping, a technique for approximating shadows based on rendering from the perspective of a light, can be combined with existing differentiable rasterizers to yield differentiable visibility information. We demonstrate at several inverse graphics problems that differentiable shadow maps are orders of magnitude faster than differentiable light transport simulation with similar accuracy -- while differentiable rasterization without shadows often fails to converge.

Rendering novel view images is highly desirable for many applications. Despite recent progress, it remains challenging to render high-fidelity and view-consistent novel views of large-scale scenes from in-the-wild images with inevitable artifacts (e.g., motion blur). To this end, we develop a hybrid neural rendering model that makes image-based representation and neural 3D representation join forces to render high-quality, view-consistent images. Besides, images captured in the wild inevitably contain artifacts, such as motion blur, which deteriorates the quality of rendered images. Accordingly, we propose strategies to simulate blur effects on the rendered images to mitigate the negative influence of blurriness images and reduce their importance during training based on precomputed quality-aware weights. Extensive experiments on real and synthetic data demonstrate our model surpasses state-of-the-art point-based methods for novel view synthesis. The code is available at https://daipengwa.github.io/Hybrid-Rendering-ProjectPage.

We propose TensoIR, a novel inverse rendering approach based on tensor factorization and neural fields. Unlike previous works that use purely MLP-based neural fields, thus suffering from low capacity and high computation costs, we extend TensoRF, a state-of-the-art approach for radiance field modeling, to estimate scene geometry, surface reflectance, and environment illumination from multi-view images captured under unknown lighting conditions. Our approach jointly achieves radiance field reconstruction and physically-based model estimation, leading to photo-realistic novel view synthesis and relighting. Benefiting from the efficiency and extensibility of the TensoRF-based representation, our method can accurately model secondary shading effects (like shadows and indirect lighting) and generally support input images captured under a single or multiple unknown lighting conditions. The low-rank tensor representation allows us to not only achieve fast and compact reconstruction but also better exploit shared information under an arbitrary number of capturing lighting conditions. We demonstrate the superiority of our method to baseline methods qualitatively and quantitatively on various challenging synthetic and real-world scenes.

By supervising camera rays between a scene and multi-view image planes, NeRF reconstructs a neural scene representation for the task of novel view synthesis. On the other hand, shadow rays between the light source and the scene have yet to be considered. Therefore, we propose a novel shadow ray supervision scheme that optimizes both the samples along the ray and the ray location. By supervising shadow rays, we successfully reconstruct a neural SDF of the scene from single-view images under multiple lighting conditions. Given single-view binary shadows, we train a neural network to reconstruct a complete scene not limited by the camera’s line of sight. By further modeling the correlation between the image colors and the shadow rays, our technique can also be effectively extended to RGB inputs. We compare our method with previous works on challenging tasks of shape reconstruction from single-view binary shadow or RGB images and observe significant improvements. The code and data are available at https://github.com/gerwang/ShadowNeuS.

Common editing operations performed by professional photographers include the cleanup operations: de-emphasizing distracting elements and enhancing subjects. These edits are challenging, requiring a delicate balance between manipulating the viewer’s attention while maintaining photo realism. While recent approaches can boast successful examples of attention attenuation or amplification, most of them also suffer from frequent unrealistic edits. We propose a realism loss for saliency-guided image enhancement to maintain high realism across varying image types, while attenuating distractors and amplifying objects of interest. Evaluations with professional photographers confirm that we achieve the dual objective of realism and effectiveness, and outperform the recent approaches on their own datasets, while requiring a smaller memory footprint and runtime. We thus offer a viable solution for automating image enhancement and photo cleanup operations.

Recent portrait relighting methods have achieved realistic results of portrait lighting effects given a desired lighting representation such as an environment map. However, these methods are not intuitive for user interaction and lack precise lighting control. We introduce LightPainter, a scribble-based relighting system that allows users to interactively manipulate portrait lighting effect with ease. This is achieved by two conditional neural networks, a delighting module that recovers geometry and albedo optionally conditioned on skin tone, and a scribble-based module for relighting. To train the relighting module, we propose a novel scribble simulation procedure to mimic real user scribbles, which allows our pipeline to be trained without any human annotations. We demonstrate high-quality and flexible portrait lighting editing capability with both quantitative and qualitative experiments. User study comparisons with commercial lighting editing tools also demonstrate consistent user preference for our method.

We propose a unified structured light, consisting of an LED array and an LCD mask, for high-quality acquisition of both shape and reflectance from a single view. For geometry, one LED projects a set of learned mask patterns to accurately encode spatial information; the decoded results from multiple LEDs are then aggregated to produce a final depth map. For appearance, learned light patterns are cast through a transparent mask to efficiently probe angularly-varying reflectance. Per-point BRDF parameters are differentiably optimized with respect to corresponding measurements, and stored in texture maps as the final reflectance. We establish a differentiable pipeline for the joint capture to automatically optimize both the mask and light patterns towards optimal acquisition quality. The effectiveness of our light is demonstrated with a wide variety of physical objects. Our results compare favorably with state-of-the-art techniques.

Detailed 3D reconstruction and photo-realistic relighting of digital humans are essential for various applications. To this end, we propose a novel sparse-view 3d human reconstruction framework that closely incorporates the occupancy field and albedo field with an additional visibility field--it not only resolves occlusion ambiguity in multiview feature aggregation, but can also be used to evaluate light attenuation for self-shadowed relighting. To enhance its training viability and efficiency, we discretize visibility onto a fixed set of sample directions and supply it with coupled geometric 3D depth feature and local 2D image feature. We further propose a novel rendering-inspired loss, namely TransferLoss, to implicitly enforce the alignment between visibility and occupancy field, enabling end-to-end joint training. Results and extensive experiments demonstrate the effectiveness of the proposed method, as it surpasses state-of-the-art in terms of reconstruction accuracy while achieving comparably accurate relighting to ray-traced ground truth.

Analyzing the dynamic changes of cellular morphology is important for understanding the various functions and characteristics of live cells, including stem cells and metastatic cancer cells. To this end, we need to track all points on the highly deformable cellular contour in every frame of live cell video. Local shapes and textures on the contour are not evident, and their motions are complex, often with expansion and contraction of local contour features. The prior arts for optical flow or deep point set tracking are unsuited due to the fluidity of cells, and previous deep contour tracking does not consider point correspondence. We propose the first deep learning-based tracking of cellular (or more generally viscoelastic materials) contours with point correspondence by fusing dense representation between two contours with cross attention. Since it is impractical to manually label dense tracking points on the contour, unsupervised learning comprised of the mechanical and cyclical consistency losses is proposed to train our contour tracker. The mechanical loss forcing the points to move perpendicular to the contour effectively helps out. For quantitative evaluation, we labeled sparse tracking points along the contour of live cells from two live cell datasets taken with phase contrast and confocal fluorescence …

Multi-view shape reconstruction has achieved impressive progresses thanks to the latest advances in neural implicit surface rendering. However, existing methods based on signed distance function (SDF) are limited to closed surfaces, failing to reconstruct a wide range of real-world objects that contain open-surface structures. In this work, we introduce a new neural rendering framework, coded NeUDF, that can reconstruct surfaces with arbitrary topologies solely from multi-view supervision. To gain the flexibility of representing arbitrary surfaces, NeUDF leverages the unsigned distance function (UDF) as surface representation. While a naive extension of SDF-based neural renderer cannot scale to UDF, we propose two new formulations of weight function specially tailored for UDF-based volume rendering. Furthermore, to cope with open surface rendering, where the in/out test is no longer valid, we present a dedicated normal regularization strategy to resolve the surface orientation ambiguity. We extensively evaluate our method over a number of challenging datasets, including DTU, MGN, and Deep Fashion 3D. Experimental results demonstrate that NeUDF can significantly outperform the state-of-the-art method in the task of multi-view surface reconstruction, especially for the complex shapes with open boundaries.

Recent progress in neural implicit functions has set new state-of-the-art in reconstructing high-fidelity 3D shapes from a collection of images. However, these approaches are limited to closed surfaces as they require the surface to be represented by a signed distance field. In this paper, we propose NeAT, a new neural rendering framework that can learn implicit surfaces with arbitrary topologies from multi-view images. In particular, NeAT represents the 3D surface as a level set of a signed distance function (SDF) with a validity branch for estimating the surface existence probability at the query positions. We also develop a novel neural volume rendering method, which uses SDF and validity to calculate the volume opacity and avoids rendering points with low validity. NeAT supports easy field-to-mesh conversion using the classic Marching Cubes algorithm. Extensive experiments on DTU, MGN, and Deep Fashion 3D datasets indicate that our approach is able to faithfully reconstruct both watertight and non-watertight surfaces. In particular, NeAT significantly outperforms the state-of-the-art methods in the task of open surface reconstruction both quantitatively and qualitatively.

This work introduces alternating latent topologies (ALTO) for high-fidelity reconstruction of implicit 3D surfaces from noisy point clouds. Previous work identifies that the spatial arrangement of latent encodings is important to recover detail. One school of thought is to encode a latent vector for each point (point latents). Another school of thought is to project point latents into a grid (grid latents) which could be a voxel grid or triplane grid. Each school of thought has tradeoffs. Grid latents are coarse and lose high-frequency detail. In contrast, point latents preserve detail. However, point latents are more difficult to decode into a surface, and quality and runtime suffer. In this paper, we propose ALTO to sequentially alternate between geometric representations, before converging to an easy-to-decode latent. We find that this preserves spatial expressiveness and makes decoding lightweight. We validate ALTO on implicit 3D recovery and observe not only a performance improvement over the state-of-the-art, but a runtime improvement of 3-10×. Anonymized source code at https://visual.ee.ucla.edu/alto.htm/.

Mesh generation is of great value in various applications involving computer graphics and virtual content, yet designing generative models for meshes is challenging due to their irregular data structure and inconsistent topology of meshes in the same category. In this work, we design a novel sparse latent point diffusion model for mesh generation. Our key insight is to regard point clouds as an intermediate representation of meshes, and model the distribution of point clouds instead. While meshes can be generated from point clouds via techniques like Shape as Points (SAP), the challenges of directly generating meshes can be effectively avoided. To boost the efficiency and controllability of our mesh generation method, we propose to further encode point clouds to a set of sparse latent points with point-wise semantic meaningful features, where two DDPMs are trained in the space of sparse latent points to respectively model the distribution of the latent point positions and features at these latent points. We find that sampling in this latent space is faster than directly sampling dense point clouds. Moreover, the sparse latent points also enable us to explicitly control both the overall structures and local details of the generated meshes. Extensive experiments are conducted …

We introduce Power Bundle Adjustment as an expansion type algorithm for solving large-scale bundle adjustment problems. It is based on the power series expansion of the inverse Schur complement and constitutes a new family of solvers that we call inverse expansion methods. We theoretically justify the use of power series and we prove the convergence of our approach. Using the real-world BAL dataset we show that the proposed solver challenges the state-of-the-art iterative methods and significantly accelerates the solution of the normal equation, even for reaching a very high accuracy. This easy-to-implement solver can also complement a recently presented distributed bundle adjustment framework. We demonstrate that employing the proposed Power Bundle Adjustment as a sub-problem solver significantly improves speed and accuracy of the distributed optimization.

We present Neural Pixel Composition (NPC), a novel approach for continuous 3D-4D view synthesis given only a discrete set of multi-view observations as input. Existing state-of-the-art approaches require dense multi-view supervision and an extensive computational budget. The proposed formulation reliably operates on sparse and wide-baseline multi-view imagery and can be trained efficiently within a few seconds to 10 minutes for hi-res (12MP) content, i.e., 200-400X faster convergence than existing methods. Crucial to our approach are two core novelties: 1) a representation of a pixel that contains color and depth information accumulated from multi-views for a particular location and time along a line of sight, and 2) a multi-layer perceptron (MLP) that enables the composition of this rich information provided for a pixel location to obtain the final color output. We experiment with a large variety of multi-view sequences, compare to existing approaches, and achieve better results in diverse and challenging settings.

Recently, DreamFusion demonstrated the utility of a pretrained text-to-image diffusion model to optimize Neural Radiance Fields (NeRF), achieving remarkable text-to-3D synthesis results. However, the method has two inherent limitations: 1) optimization of the NeRF representation is extremely slow, 2) NeRF is supervised by images at a low resolution (64×64), thus leading to low-quality 3D models with a long wait time. In this paper, we address these limitations by utilizing a two-stage coarse-to-fine optimization framework. In the first stage, we use a sparse 3D neural representation to accelerate optimization while using a low-resolution diffusion prior. In the second stage, we use a textured mesh model initialized from the coarse neural representation, allowing us to perform optimization with a very efficient differentiable renderer interacting with high-resolution images. Our method, dubbed Magic3D, can create a 3D mesh model in 40 minutes, 2× faster than DreamFusion (reportedly taking 1.5 hours on average), while achieving 8× higher resolution. User studies show 61.7% raters to prefer our approach than DreamFusion. Together with the image-conditioned generation capabilities, we provide users with new ways to control 3D synthesis, opening up new avenues to various creative applications.

Looping videos are short video clips that can be looped endlessly without visible seams or artifacts. They provide a very attractive way to capture the dynamism of natural scenes. Existing methods have been mostly limited to 2D representations. In this paper, we take a step forward and propose a practical solution that enables an immersive experience on dynamic 3D looping scenes. The key challenge is to consider the per-view looping conditions from asynchronous input while maintaining view consistency for the 3D representation. We propose a novel sparse 3D video representation, namely Multi-Tile Video (MTV), which not only provides a view-consistent prior, but also greatly reduces memory usage, making the optimization of a 4D volume tractable. Then, we introduce a two-stage pipeline to construct the 3D looping MTV from completely asynchronous multi-view videos with no time overlap. A novel looping loss based on video temporal retargeting algorithms is adopted during the optimization to loop the 3D scene. Experiments of our framework have shown promise in successfully generating and rendering photorealistic 3D looping videos in real time even on mobile devices. The code, dataset, and live demos are available in https://limacv.github.io/VideoLoop3D_web/.

We present a high-fidelity 3D generative adversarial network (GAN) inversion framework that can synthesize photo-realistic novel views while preserving specific details of the input image. High-fidelity 3D GAN inversion is inherently challenging due to the geometry-texture trade-off, where overfitting to a single view input image often damages the estimated geometry during the latent optimization. To solve this challenge, we propose a novel pipeline that builds on the pseudo-multi-view estimation with visibility analysis. We keep the original textures for the visible parts and utilize generative priors for the occluded parts. Extensive experiments show that our approach achieves advantageous reconstruction and novel view synthesis quality over prior work, even for images with out-of-distribution textures. The proposed pipeline also enables image attribute editing with the inverted latent code and 3D-aware texture modification. Our approach enables high-fidelity 3D rendering from a single image, which is promising for various applications of AI-generated 3D content. The source code is at https://github.com/jiaxinxie97/HFGI3D/.

This work explores the use of 3D generative models to synthesize training data for 3D vision tasks. The key requirements of the generative models are that the generated data should be photorealistic to match the real-world scenarios, and the corresponding 3D attributes should be aligned with given sampling labels. However, we find that the recent NeRF-based 3D GANs hardly meet the above requirements due to their designed generation pipeline and the lack of explicit 3D supervision. In this work, we propose Lift3D, an inverted 2D-to-3D generation framework to achieve the data generation objectives. Lift3D has several merits compared to prior methods: (1) Unlike previous 3D GANs that the output resolution is fixed after training, Lift3D can generalize to any camera intrinsic with higher resolution and photorealistic output. (2) By lifting well-disentangled 2D GAN to 3D object NeRF, Lift3D provides explicit 3D information of generated objects, thus offering accurate 3D annotations for downstream tasks. We evaluate the effectiveness of our framework by augmenting autonomous driving datasets. Experimental results demonstrate that our data generation framework can effectively improve the performance of 3D object detectors. Code: len-li.github.io/lift3d-web

Recently, a surge of high-quality 3D-aware GANs have been proposed, which leverage the generative power of neural rendering. It is natural to associate 3D GANs with GAN inversion methods to project a real image into the generator’s latent space, allowing free-view consistent synthesis and editing, referred as 3D GAN inversion. Although with the facial prior preserved in pre-trained 3D GANs, reconstructing a 3D portrait with only one monocular image is still an ill-pose problem. The straightforward application of 2D GAN inversion methods focuses on texture similarity only while ignoring the correctness of 3D geometry shapes. It may raise geometry collapse effects, especially when reconstructing a side face under an extreme pose. Besides, the synthetic results in novel views are prone to be blurry. In this work, we propose a novel method to promote 3D GAN inversion by introducing facial symmetry prior. We design a pipeline and constraints to make full use of the pseudo auxiliary view obtained via image flipping, which helps obtain a view-consistent and well-structured geometry shape during the inversion process. To enhance texture fidelity in unobserved viewpoints, pseudo labels from depth-guided 3D warping can provide extra supervision. We design constraints aimed at filtering out conflict areas for …

Recent researches reveal that StyleGAN can generate highly realistic images, inspiring researchers to use pretrained StyleGAN to generate high-fidelity swapped faces. However, existing methods fail to meet the expectations in two essential aspects of high-fidelity face swapping. Their results are blurry without pore-level details and fail to preserve identity for challenging cases. To overcome the above artifacts, we innovatively construct a series of identity-preserving semantic bases of StyleGAN (called StyleIPSB) in respect of pose, expression, and illumination. Each basis of StyleIPSB controls one specific semantic attribute and disentangles with the others. The StyleIPSB constrains style code in the subspace of W+ space to preserve pore-level details. StyleIPSB gives us a novel tool for high-fidelity face swapping, and we propose a three-stage framework for face swapping with StyleIPSB. Firstly, we transform the target facial images’ attributes to the source image. We learn the mapping from 3D Morphable Model (3DMM) parameters, which capture the prominent semantic variance, to the coordinates of StyleIPSB that show higher identity-preserving and fidelity. Secondly, to transform detailed attributes which 3DMM does not capture, we learn the residual attribute between the reenacted face and the target face. Finally, the face is blended into the background of the target …

We present a large-scale facial UV-texture dataset that contains over 50,000 high-quality texture UV-maps with even illuminations, neutral expressions, and cleaned facial regions, which are desired characteristics for rendering realistic 3D face models under different lighting conditions. The dataset is derived from a large-scale face image dataset namely FFHQ, with the help of our fully automatic and robust UV-texture production pipeline. Our pipeline utilizes the recent advances in StyleGAN-based facial image editing approaches to generate multi-view normalized face images from single-image inputs. An elaborated UV-texture extraction, correction, and completion procedure is then applied to produce high-quality UV-maps from the normalized face images. Compared with existing UV-texture datasets, our dataset has more diverse and higher-quality texture maps. We further train a GAN-based texture decoder as the nonlinear texture basis for parametric fitting based 3D face reconstruction. Experiments show that our method improves the reconstruction accuracy over state-of-the-art approaches, and more importantly, produces high-quality texture maps that are ready for realistic renderings. The dataset, code, and pre-trained texture decoder are publicly available at https://github.com/csbhr/FFHQ-UV.

In this work, we aim to enhance model-based face reconstruction by avoiding fitting the model to outliers, i.e. regions that cannot be well-expressed by the model such as occluders or make-up. The core challenge for localizing outliers is that they are highly variable and difficult to annotate. To overcome this challenging problem, we introduce a joint Face-autoencoder and outlier segmentation approach (FOCUS).In particular, we exploit the fact that the outliers cannot be fitted well by the face model and hence can be localized well given a high-quality model fitting. The main challenge is that the model fitting and the outlier segmentation are mutually dependent on each other, and need to be inferred jointly. We resolve this chicken-and-egg problem with an EM-type training strategy, where a face autoencoder is trained jointly with an outlier segmentation network. This leads to a synergistic effect, in which the segmentation network prevents the face encoder from fitting to the outliers, enhancing the reconstruction quality. The improved 3D face reconstruction, in turn, enables the segmentation network to better predict the outliers. To resolve the ambiguity between outliers and regions that are difficult to fit, such as eyebrows, we build a statistical prior from synthetic data that …

Generative models show good potential for recovering 3D faces beyond limited shape assumptions. While plausible details and resolutions are achieved, these models easily fail under extreme conditions of pose, shadow or appearance, due to the entangled fitting or lack of multi-view priors. To address this problem, this paper presents a novel Neural Proto-face Field (NPF) for unsupervised robust 3D face modeling. Instead of using constrained images as Neural Radiance Field (NeRF), NPF disentangles the common/specific facial cues, i.e., ID, expression and scene-specific details from in-the-wild photo collections. Specifically, NPF learns a face prototype to aggregate 3D-consistent identity via uncertainty modeling, extracting multi-image priors from a photo collection. NPF then learns to deform the prototype with the appropriate facial expressions, constrained by a loss of expression consistency and personal idiosyncrasies. Finally, NPF is optimized to fit a target image in the collection, recovering specific details of appearance and geometry. In this way, the generative model benefits from multi-image priors and meaningful facial structures. Extensive experiments on benchmarks show that NPF recovers superior or competitive facial shapes and textures, compared to state-of-the-art methods.

Limited by the nature of the low-dimensional representational capacity of 3DMM, most of the 3DMM-based face reconstruction (FR) methods fail to recover high-frequency facial details, such as wrinkles, dimples, etc. Some attempt to solve the problem by introducing detail maps or non-linear operations, however, the results are still not vivid. To this end, we in this paper present a novel hierarchical representation network (HRN) to achieve accurate and detailed face reconstruction from a single image. Specifically, we implement the geometry disentanglement and introduce the hierarchical representation to fulfill detailed face modeling. Meanwhile, 3D priors of facial details are incorporated to enhance the accuracy and authenticity of the reconstruction results. We also propose a de-retouching module to achieve better decoupling of the geometry and appearance. It is noteworthy that our framework can be extended to a multi-view fashion by considering detail consistency of different views. Extensive experiments on two single-view and two multi-view FR benchmarks demonstrate that our method outperforms the existing methods in both reconstruction accuracy and visual effects. Finally, we introduce a high-quality 3D face dataset FaceHD-100 to boost the research of high-fidelity face reconstruction. The project homepage is at https://younglbw.github.io/HRN-homepage/.

Generating faithful visualizations of human faces requires capturing both coarse and fine-level details of the face geometry and appearance. Existing methods are either data-driven, requiring an extensive corpus of data not publicly accessible to the research community, or fail to capture fine details because they rely on geometric face models that cannot represent fine-grained details in texture with a mesh discretization and linear deformation designed to model only a coarse face geometry. We introduce a method that bridges this gap by drawing inspiration from traditional computer graphics techniques. Unseen expressions are modeled by blending appearance from a sparse set of extreme poses. This blending is performed by measuring local volumetric changes in those expressions and locally reproducing their appearance whenever a similar expression is performed at test time. We show that our method generalizes to unseen expressions, adding fine-grained effects on top of smooth volumetric deformations of a face, and demonstrate how it generalizes beyond faces.

High-quality reconstruction of controllable 3D head avatars from 2D videos is highly desirable for virtual human applications in movies, games, and telepresence. Neural implicit fields provide a powerful representation to model 3D head avatars with personalized shape, expressions, and facial parts, e.g., hair and mouth interior, that go beyond the linear 3D morphable model (3DMM). However, existing methods do not model faces with fine-scale facial features, or local control of facial parts that extrapolate asymmetric expressions from monocular videos. Further, most condition only on 3DMM parameters with poor(er) locality, and resolve local features with a global neural field. We build on part-based implicit shape models that decompose a global deformation field into local ones. Our novel formulation models multiple implicit deformation fields with local semantic rig-like control via 3DMM-based parameters, and representative facial landmarks. Further, we propose a local control loss and attention mask mechanism that promote sparsity of each learned deformation field. Our formulation renders sharper locally controllable nonlinear deformations than previous implicit monocular approaches, especially mouth interior, asymmetric expressions, and facial details. Project page:https://imaging.cs.cmu.edu/localdeformationfields/

One-shot video-driven talking face generation aims at producing a synthetic talking video by transferring the facial motion from a video to an arbitrary portrait image. Head pose and facial expression are always entangled in facial motion and transferred simultaneously. However, the entanglement sets up a barrier for these methods to be used in video portrait editing directly, where it may require to modify the expression only while maintaining the pose unchanged. One challenge of decoupling pose and expression is the lack of paired data, such as the same pose but different expressions. Only a few methods attempt to tackle this challenge with the feat of 3D Morphable Models (3DMMs) for explicit disentanglement. But 3DMMs are not accurate enough to capture facial details due to the limited number of Blendshapes, which has side effects on motion transfer. In this paper, we introduce a novel self-supervised disentanglement framework to decouple pose and expression without 3DMMs and paired data, which consists of a motion editing module, a pose generator, and an expression generator. The editing module projects faces into a latent space where pose motion and expression motion can be disentangled, and the pose or expression transfer can be performed in the latent …

To bring digital avatars into people’s lives, it is highly demanded to efficiently generate complete, realistic, and animatable head avatars. This task is challenging, and it is difficult for existing methods to satisfy all the requirements at once. To achieve these goals, we propose GANHead (Generative Animatable Neural Head Avatar), a novel generative head model that takes advantages of both the fine-grained control over the explicit expression parameters and the realistic rendering results of implicit representations. Specifically, GANHead represents coarse geometry, fine-gained details and texture via three networks in canonical space to obtain the ability to generate complete and realistic head avatars. To achieve flexible animation, we define the deformation filed by standard linear blend skinning (LBS), with the learned continuous pose and expression bases and LBS weights. This allows the avatars to be directly animated by FLAME parameters and generalize well to unseen poses and expressions. Compared to state-of-the-art (SOTA) methods, GANHead achieves superior performance on head avatar generation and raw scan fitting.

Dance is an important human art form, but creating new dances can be difficult and time-consuming. In this work, we introduce Editable Dance GEneration (EDGE), a state-of-the-art method for editable dance generation that is capable of creating realistic, physically-plausible dances while remaining faithful to the input music. EDGE uses a transformer-based diffusion model paired with Jukebox, a strong music feature extractor, and confers powerful editing capabilities well-suited to dance, including joint-wise conditioning, and in-betweening. We introduce a new metric for physical plausibility, and evaluate dance quality generated by our method extensively through (1) multiple quantitative metrics on physical plausibility, alignment, and diversity benchmarks, and more importantly, (2) a large-scale user study, demonstrating a significant improvement over previous state-of-the-art methods. Qualitative samples from our model can be found at our website.

Cinemagraphs are short looping videos created by adding subtle motions to a static image. This kind of media is popular and engaging. However, automatic generation of cinemagraphs is an underexplored area and current solutions require tedious low-level manual authoring by artists. In this paper, we present an automatic method that allows generating human cinemagraphs from single RGB images. We investigate the problem in the context of dressed humans under the wind. At the core of our method is a novel cyclic neural network that produces looping cinemagraphs for the target loop duration. To circumvent the problem of collecting real data, we demonstrate that it is possible, by working in the image normal space, to learn garment motion dynamics on synthetic data and generalize to real data. We evaluate our method on both synthetic and real data and demonstrate that it is possible to create compelling and plausible cinemagraphs from single RGB images.

This work addresses the problem of generating 3D holistic body motions from human speech. Given a speech recording, we synthesize sequences of 3D body poses, hand gestures, and facial expressions that are realistic and diverse. To achieve this, we first build a high-quality dataset of 3D holistic body meshes with synchronous speech. We then define a novel speech-to-motion generation framework in which the face, body, and hands are modeled separately. The separated modeling stems from the fact that face articulation strongly correlates with human speech, while body poses and hand gestures are less correlated. Specifically, we employ an autoencoder for face motions, and a compositional vector-quantized variational autoencoder (VQ-VAE) for the body and hand motions. The compositional VQ-VAE is key to generating diverse results. Additionally, we propose a cross conditional autoregressive model that generates body poses and hand gestures, leading to coherent and realistic motions. Extensive experiments and user studies demonstrate that our proposed approach achieves state-of-the-art performance both qualitatively and quantitatively. Our novel dataset and code will be released for research purposes.

With the recent surge in popularity of AR/VR applications, realistic and accurate control of 3D full-body avatars has become a highly demanded feature. A particular challenge is that only a sparse tracking signal is available from standalone HMDs (Head Mounted Devices), often limited to tracking the user’s head and wrists. While this signal is resourceful for reconstructing the upper body motion, the lower body is not tracked and must be synthesized from the limited information provided by the upper body joints. In this paper, we present AGRoL, a novel conditional diffusion model specifically designed to track full bodies given sparse upper-body tracking signals. Our model is based on a simple multi-layer perceptron (MLP) architecture and a novel conditioning scheme for motion data. It can predict accurate and smooth full-body motion, particularly the challenging lower body movement. Unlike common diffusion architectures, our compact architecture can run in real-time, making it suitable for online body-tracking applications. We train and evaluate our model on AMASS motion capture dataset, and demonstrate that our approach outperforms state-of-the-art methods in generated motion accuracy and smoothness. We further justify our design choices through extensive experiments and ablation studies.

This paper proposes an anchor-based deformation model, namely AnchorDEF, to predict 3D garment animation from a body motion sequence. It deforms a garment mesh template by a mixture of rigid transformations with extra nonlinear displacements. A set of anchors around the mesh surface is introduced to guide the learning of rigid transformation matrices. Once the anchor transformations are found, per-vertex nonlinear displacements of the garment template can be regressed in a canonical space, which reduces the complexity of deformation space learning. By explicitly constraining the transformed anchors to satisfy the consistencies of position, normal and direction, the physical meaning of learned anchor transformations in space is guaranteed for better generalization. Furthermore, an adaptive anchor updating is proposed to optimize the anchor position by being aware of local mesh topology for learning representative anchor transformations. Qualitative and quantitative experiments on different types of garments demonstrate that AnchorDEF achieves the state-of-the-art performance on 3D garment deformation prediction in motion, especially for loose-fitting garments.

Creating animatable avatars from static scans requires the modeling of clothing deformations in different poses. Existing learning-based methods typically add pose-dependent deformations upon a minimally-clothed mesh template or a learned implicit template, which have limitations in capturing details or hinder end-to-end learning. In this paper, we revisit point-based solutions and propose to decompose explicit garment-related templates and then add pose-dependent wrinkles to them. In this way, the clothing deformations are disentangled such that the pose-dependent wrinkles can be better learned and applied to unseen poses. Additionally, to tackle the seam artifact issues in recent state-of-the-art point-based methods, we propose to learn point features on a body surface, which establishes a continuous and compact feature space to capture the fine-grained and pose-dependent clothing geometry. To facilitate the research in this field, we also introduce a high-quality scan dataset of humans in real-world clothing. Our approach is validated on two existing datasets and our newly introduced dataset, showing better clothing deformation results in unseen poses. The project page with code and dataset can be found at https://www.liuyebin.com/closet.

The combination of deep learning, artist-curated scans, and Implicit Functions (IF), is enabling the creation of detailed, clothed, 3D humans from images. However, existing methods are far from perfect. IF-based methods recover free-form geometry, but produce disembodied limbs or degenerate shapes for novel poses or clothes. To increase robustness for these cases, existing work uses an explicit parametric body model to constrain surface reconstruction, but this limits the recovery of free-form surfaces such as loose clothing that deviates from the body. What we want is a method that combines the best properties of implicit representation and explicit body regularization. To this end, we make two key observations: (1) current networks are better at inferring detailed 2D maps than full-3D surfaces, and (2) a parametric model can be seen as a “canvas” for stitching together detailed surface patches. Based on these, our method, ECON, has three main steps: (1) It infers detailed 2D normal maps for the front and back side of a clothed person. (2) From these, it recovers 2.5D front and back surfaces, called d-BiNI, that are equally detailed, yet incomplete, and registers these w.r.t. each other with the help of a SMPL-X body mesh recovered from the image. …

We present PersonNeRF, a method that takes a collection of photos of a subject (e.g., Roger Federer) captured across multiple years with arbitrary body poses and appearances, and enables rendering the subject with arbitrary novel combinations of viewpoint, body pose, and appearance. PersonNeRF builds a customized neural volumetric 3D model of the subject that is able to render an entire space spanned by camera viewpoint, body pose, and appearance. A central challenge in this task is dealing with sparse observations; a given body pose is likely only observed by a single viewpoint with a single appearance, and a given appearance is only observed under a handful of different body poses. We address this issue by recovering a canonical T-pose neural volumetric representation of the subject that allows for changing appearance across different observations, but uses a shared pose-dependent motion field across all observations. We demonstrate that this approach, along with regularization of the recovered volumetric geometry to encourage smoothness, is able to recover a model that renders compelling images from novel combinations of viewpoint, pose, and appearance from these challenging unstructured photo collections, outperforming prior work for free-viewpoint human rendering.

Inspired by the success of volumetric 3D pose estimation, some recent human mesh estimators propose to estimate 3D skeletons as intermediate representations, from which, the dense 3D meshes are regressed by exploiting the mesh topology. However, body shape information is lost in extracting skeletons, leading to mediocre performance. The advanced motion capture systems solve the problem by placing dense physical markers on the body surface, which allows to extract realistic meshes from their non-rigid motions. However, they cannot be applied to wild images without markers. In this work, we present an intermediate representation, named virtual markers, which learns 64 landmark keypoints on the body surface based on the large-scale mocap data in a generative style, mimicking the effects of physical markers. The virtual markers can be accurately detected from wild images and can reconstruct the intact meshes with realistic shapes by simple interpolation. Our approach outperforms the state-of-the-art methods on three datasets. In particular, it surpasses the existing methods by a notable margin on the SURREAL dataset, which has diverse body shapes. Code is available at https://github.com/ShirleyMaxx/VirtualMarker.

Direct mesh fitting for 3D hand shape reconstruction estimates highly accurate meshes. However, the resulting meshes are prone to artifacts and do not appear as plausible hand shapes. Conversely, parametric models like MANO ensure plausible hand shapes but are not as accurate as the non-parametric methods. In this work, we introduce a novel weakly-supervised hand shape estimation framework that integrates non-parametric mesh fitting with MANO models in an end-to-end fashion. Our joint model overcomes the tradeoff in accuracy and plausibility to yield well-aligned and high-quality 3D meshes, especially in challenging two-hand and hand-object interaction scenarios.

Hands, one of the most dynamic parts of our body, suffer from blur due to their active movements. However, previous 3D hand mesh recovery methods have mainly focused on sharp hand images rather than considering blur due to the absence of datasets providing blurry hand images. We first present a novel dataset BlurHand, which contains blurry hand images with 3D groundtruths. The BlurHand is constructed by synthesizing motion blur from sequential sharp hand images, imitating realistic and natural motion blurs. In addition to the new dataset, we propose BlurHandNet, a baseline network for accurate 3D hand mesh recovery from a blurry hand image. Our BlurHandNet unfolds a blurry input image to a 3D hand mesh sequence to utilize temporal information in the blurry input image, while previous works output a static single hand mesh. We demonstrate the usefulness of BlurHand for the 3D hand mesh recovery from blurry images in our experiments. The proposed BlurHandNet produces much more robust results on blurry images while generalizing well to in-the-wild images. The training codes and BlurHand dataset are available at https://github.com/JaehaKim97/BlurHand_RELEASE.

Existing methods proposed for hand reconstruction tasks usually parameterize a generic 3D hand model or predict hand mesh positions directly. The parametric representations consisting of hand shapes and rotational poses are more stable, while the non-parametric methods can predict more accurate mesh positions. In this paper, we propose to reconstruct meshes and estimate MANO parameters of two hands from a single RGB image simultaneously to utilize the merits of two kinds of hand representations. To fulfill this target, we propose novel Mesh-Mano interaction blocks (MMIBs), which take mesh vertices positions and MANO parameters as two kinds of query tokens. MMIB consists of one graph residual block to aggregate local information and two transformer encoders to model long-range dependencies. The transformer encoders are equipped with different asymmetric attention masks to model the intra-hand and inter-hand attention, respectively. Moreover, we introduce the mesh alignment refinement module to further enhance the mesh-image alignment. Extensive experiments on the InterHand2.6M benchmark demonstrate promising results over the state-of-the-art hand reconstruction methods.

We introduce PLIKS (Pseudo-Linear Inverse Kinematic Solver) for reconstruction of a 3D mesh of the human body from a single 2D image. Current techniques directly regress the shape, pose, and translation of a parametric model from an input image through a non-linear mapping with minimal flexibility to any external influences. We approach the task as a model-in-the-loop optimization problem. PLIKS is built on a linearized formulation of the parametric SMPL model. Using PLIKS, we can analytically reconstruct the human model via 2D pixel-aligned vertices. This enables us with the flexibility to use accurate camera calibration information when available. PLIKS offers an easy way to introduce additional constraints such as shape and translation. We present quantitative evaluations which confirm that PLIKS achieves more accurate reconstruction with greater than 10% improvement compared to other state-of-the-art methods with respect to the standard 3D human pose and shape benchmarks while also obtaining a reconstruction error improvement of 12.9 mm on the newer AGORA dataset.

In this work, we focus on a novel task of category-level functional hand-object manipulation synthesis covering both rigid and articulated object categories. Given an object geometry, an initial human hand pose as well as a sparse control sequence of object poses, our goal is to generate a physically reasonable hand-object manipulation sequence that performs like human beings. To address such a challenge, we first design CAnonicalized Manipulation Spaces (CAMS), a two-level space hierarchy that canonicalizes the hand poses in an object-centric and contact-centric view. Benefiting from the representation capability of CAMS, we then present a two-stage framework for synthesizing human-like manipulation animations. Our framework achieves state-of-the-art performance for both rigid and articulated categories with impressive visual effects. Codes and video results can be found at our project homepage: https://cams-hoi.github.io/.

Convenient 4D modeling of human-object interactions is essential for numerous applications. However, monocular tracking and rendering of complex interaction scenarios remain challenging. In this paper, we propose Instant-NVR, a neural approach for instant volumetric human-object tracking and rendering using a single RGBD camera. It bridges traditional non-rigid tracking with recent instant radiance field techniques via a multi-thread tracking-rendering mechanism. In the tracking front-end, we adopt a robust human-object capture scheme to provide sufficient motion priors. We further introduce a separated instant neural representation with a novel hybrid deformation module for the interacting scene. We also provide an on-the-fly reconstruction scheme of the dynamic/static radiance fields via efficient motion-prior searching. Moreover, we introduce an online key frame selection scheme and a rendering-aware refinement strategy to significantly improve the appearance details for online novel-view synthesis. Extensive experiments demonstrate the effectiveness and efficiency of our approach for the instant generation of human-object radiance fields on the fly, notably achieving real-time photo-realistic novel view synthesis under complex human-object interactions.

We present a near real-time (10Hz) method for 6-DoF tracking of an unknown object from a monocular RGBD video sequence, while simultaneously performing neural 3D reconstruction of the object. Our method works for arbitrary rigid objects, even when visual texture is largely absent. The object is assumed to be segmented in the first frame only. No additional information is required, and no assumption is made about the interaction agent. Key to our method is a Neural Object Field that is learned concurrently with a pose graph optimization process in order to robustly accumulate information into a consistent 3D representation capturing both geometry and appearance. A dynamic pool of posed memory frames is automatically maintained to facilitate communication between these threads. Our approach handles challenging sequences with large pose changes, partial and full occlusion, untextured surfaces, and specular highlights. We show results on HO3D, YCBInEOAT, and BEHAVE datasets, demonstrating that our method significantly outperforms existing approaches. Project page: https://bundlesdf.github.io/

Humans have long been recorded in a variety of forms since antiquity. For example, sculptures and paintings were the primary media for depicting human beings before the invention of cameras. However, most current human-centric computer vision tasks like human pose estimation and human image generation focus exclusively on natural images in the real world. Artificial humans, such as those in sculptures, paintings, and cartoons, are commonly neglected, making existing models fail in these scenarios. As an abstraction of life, art incorporates humans in both natural and artificial scenes. We take advantage of it and introduce the Human-Art dataset to bridge related tasks in natural and artificial scenarios. Specifically, Human-Art contains 50k high-quality images with over 123k person instances from 5 natural and 15 artificial scenarios, which are annotated with bounding boxes, keypoints, self-contact points, and text information for humans represented in both 2D and 3D. It is, therefore, comprehensive and versatile for various downstream tasks. We also provide a rich set of baseline results and detailed analyses for related tasks, including human detection, 2D and 3D human pose estimation, image generation, and motion transfer. As a challenging dataset, we hope Human-Art can provide insights for relevant research and open up …

We propose a method to decompose a video into a background and a set of foreground layers, where the background captures stationary elements while the foreground layers capture moving objects along with their associated effects (e.g. shadows and reflections). Our approach is designed for unconstrained monocular videos, with arbitrary camera and object motion. Prior work that tackles this problem assumes that the video can be mapped onto a fixed 2D canvas, severely limiting the possible space of camera motion. Instead, our method applies recent progress in monocular camera pose and depth estimation to create a full, RGBD video layer for the background, along with a video layer for each foreground object. To solve the underconstrained decomposition problem, we propose a new loss formulation based on multi-view consistency. We test our method on challenging videos with complex camera motion and show significant qualitative improvement over current approaches.

In this work we study the benefits of using tracking and 3D poses for action recognition. To achieve this, we take the Lagrangian view on analysing actions over a trajectory of human motion rather than at a fixed point in space. Taking this stand allows us to use the tracklets of people to predict their actions. In this spirit, first we show the benefits of using 3D pose to infer actions, and study person-person interactions. Subsequently, we propose a Lagrangian Action Recognition model by fusing 3D pose and contextualized appearance over tracklets. To this end, our method achieves state-of-the-art performance on the AVA v2.2 dataset on both pose only settings and on standard benchmark settings. When reasoning about the action using only pose cues, our pose model achieves +10.0 mAP gain over the corresponding state-of-the-art while our fused model has a gain of +2.8 mAP over the best state-of-the-art model. Code and results are available at: https://brjathu.github.io/LART

Towards stable human pose estimation from monocular images, there remain two main dilemmas. On the one hand, the different perspectives, i.e., front view, side view, and top view, appear the inconsistent performances due to the depth ambiguity. On the other hand, foot posture plays a significant role in complicated human pose estimation, i.e., dance and sports, and foot-ground interaction, but unfortunately, it is omitted in most general approaches and datasets. In this paper, we first propose the Cross-View Fusion (CVF) module to catch up with better 3D intermediate representation and alleviate the view inconsistency based on the vision transformer encoder. Then the optimization-based method is introduced to reconstruct the foot pose and foot-ground contact for the general multi-view datasets including AIST++ and Human3.6M. Besides, the reversible kinematic topology strategy is innovated to utilize the contact information into the full-body with foot pose regressor. Extensive experiments on the popular benchmarks demonstrate that our method outperforms the state-of-the-art approaches by achieving 40.1mm PA-MPJPE on the 3DPW test set and 43.8mm on the AIST++ test set.

Human pose is typically represented by a coordinate vector of body joints or their heatmap embeddings. While easy for data processing, unrealistic pose estimates are admitted due to the lack of dependency modeling between the body joints. In this paper, we present a structured representation, named Pose as Compositional Tokens (PCT), to explore the joint dependency. It represents a pose by M discrete tokens with each characterizing a sub-structure with several interdependent joints. The compositional design enables it to achieve a small reconstruction error at a low cost. Then we cast pose estimation as a classification task. In particular, we learn a classifier to predict the categories of the M tokens from an image. A pre-learned decoder network is used to recover the pose from the tokens without further post-processing. We show that it achieves better or comparable pose estimation results as the existing methods in general scenarios, yet continues to work well when occlusion occurs, which is ubiquitous in practice. The code and models are publicly available at https://github.com/Gengzigang/PCT.

Human pose and shape (HPS) estimation methods achieve remarkable results. However, current HPS benchmarks are mostly designed to test models in scenarios that are similar to the training data. This can lead to critical situations in real-world applications when the observed data differs significantly from the training data and hence is out-of-distribution (OOD). It is therefore important to test and improve the OOD robustness of HPS methods. To address this fundamental problem, we develop a simulator that can be controlled in a fine-grained manner using interpretable parameters to explore the manifold of images of human pose, e.g. by varying poses, shapes, and clothes. We introduce a learning-based testing method, termed PoseExaminer, that automatically diagnoses HPS algorithms by searching over the parameter space of human pose images to find the failure modes. Our strategy for exploring this high-dimensional parameter space is a multi-agent reinforcement learning system, in which the agents collaborate to explore different parts of the parameter space. We show that our PoseExaminer discovers a variety of limitations in current state-of-the-art models that are relevant in real-world scenarios but are missed by current benchmarks. For example, it finds large regions of realistic human poses that are not predicted correctly, as …

We present SLOPER4D, a novel scene-aware dataset collected in large urban environments to facilitate the research of global human pose estimation (GHPE) with human-scene interaction in the wild. Employing a head-mounted device integrated with a LiDAR and camera, we record 12 human subjects’ activities over 10 diverse urban scenes from an egocentric view. Frame-wise annotations for 2D key points, 3D pose parameters, and global translations are provided, together with reconstructed scene point clouds. To obtain accurate 3D ground truth in such large dynamic scenes, we propose a joint optimization method to fit local SMPL meshes to the scene and fine-tune the camera calibration during dynamic motions frame by frame, resulting in plausible and scene-natural 3D human poses. Eventually, SLOPER4D consists of 15 sequences of human motions, each of which has a trajectory length of more than 200 meters (up to 1,300 meters) and covers an area of more than 200 square meters (up to 30,000 square meters), including more than 100k LiDAR frames, 300k video frames, and 500K IMU-based motion frames. With SLOPER4D, we provide a detailed and thorough analysis of two critical tasks, including camera-based 3D HPE and LiDAR-based 3D HPE in urban environments, and benchmark a new task, …

In this paper, we delve into semi-supervised 2D human pose estimation. The previous method ignored two problems: (i) When conducting interactive training between large model and lightweight model, the pseudo label of lightweight model will be used to guide large models. (ii) The negative impact of noise pseudo labels on training. Moreover, the labels used for 2D human pose estimation are relatively complex: keypoint category and keypoint position. To solve the problems mentioned above, we propose a semi-supervised 2D human pose estimation framework driven by a position inconsistency pseudo label correction module (SSPCM). We introduce an additional auxiliary teacher and use the pseudo labels generated by the two teacher model in different periods to calculate the inconsistency score and remove outliers. Then, the two teacher models are updated through interactive training, and the student model is updated using the pseudo labels generated by two teachers. To further improve the performance of the student model, we use the semi-supervised Cut-Occlude based on pseudo keypoint perception to generate more hard and effective samples. In addition, we also proposed a new indoor overhead fisheye human keypoint dataset WEPDTOF-Pose. Extensive experiments demonstrate that our method outperforms the previous best semi-supervised 2D human pose estimation …

We study human pose estimation in extremely low-light images. This task is challenging due to the difficulty of collecting real low-light images with accurate labels, and severely corrupted inputs that degrade prediction quality significantly. To address the first issue, we develop a dedicated camera system and build a new dataset of real low-light images with accurate pose labels. Thanks to our camera system, each low-light image in our dataset is coupled with an aligned well-lit image, which enables accurate pose labeling and is used as privileged information during training. We also propose a new model and a new training strategy that fully exploit the privileged information to learn representation insensitive to lighting conditions. Our method demonstrates outstanding performance on real extremely low-light images, and extensive analyses validate that both of our model and dataset contribute to the success.

Deep learning has been utilized in knowledge-based radiotherapy planning in which a system trained with a set of clinically approved plans is employed to infer a three-dimensional dose map for a given new patient. However, previous deep methods are primarily limited to simple scenarios, e.g., a fixed planning type or a consistent beam angle configuration. This in fact limits the usability of such approaches and makes them not generalizable over a larger set of clinical scenarios. Herein, we propose a novel conditional generative model, Flexible-C^m GAN, utilizing additional information regarding planning types and various beam geometries. A miss-consistency loss is proposed to deal with the challenge of having a limited set of conditions on the input data, e.g., incomplete training samples. To address the challenges of including clinical preferences, we derive a differentiable shift-dose-volume loss to incorporate the well-known dose-volume histogram constraints. During inference, users can flexibly choose a specific planning type and a set of beam angles to meet the clinical requirements. We conduct experiments on an illustrative face dataset to show the motivation of Flexible-C^m GAN and further validate our model’s potential clinical values with two radiotherapy datasets. The results demonstrate the superior performance of the proposed method …

Self-supervised multi-frame depth estimation achieves high accuracy by computing matching costs of pixel correspondences between adjacent frames, injecting geometric information into the network. These pixel-correspondence candidates are computed based on the relative pose estimates between the frames. Accurate pose predictions are essential for precise matching cost computation as they influence the epipolar geometry. Furthermore, improved depth estimates can, in turn, be used to align pose estimates. Inspired by traditional structure-from-motion (SfM) principles, we propose the DualRefine model, which tightly couples depth and pose estimation through a feedback loop. Our novel update pipeline uses a deep equilibrium model framework to iteratively refine depth estimates and a hidden state of feature maps by computing local matching costs based on epipolar geometry. Importantly, we used the refined depth estimates and feature maps to compute pose updates at each step. This update in the pose estimates slowly alters the epipolar geometry during the refinement process. Experimental results on the KITTI dataset demonstrate competitive depth prediction and odometry prediction performance surpassing published self-supervised baselines. The code is available at https://github.com/antabangun/DualRefine.

We propose a novel visual-inertial odometry (VIO) initialization method, which decouples rotation and translation estimation, and achieves higher efficiency and better robustness. Existing loosely-coupled VIO-initialization methods suffer from poor stability of visual structure-from-motion (SfM), whereas those tightly-coupled methods often ignore the gyroscope bias in the closed-form solution, resulting in limited accuracy. Moreover, the aforementioned two classes of methods are computationally expensive, because 3D point clouds need to be reconstructed simultaneously. In contrast, our new method fully combines inertial and visual measurements for both rotational and translational initialization. First, a rotation-only solution is designed for gyroscope bias estimation, which tightly couples the gyroscope and camera observations. Second, the initial velocity and gravity vector are solved with linear translation constraints in a globally optimal fashion and without reconstructing 3D point clouds. Extensive experiments have demonstrated that our method is 8~72 times faster (w.r.t. a 10-frame set) than the state-of-the-art methods, and also presents significantly higher robustness and accuracy. The source code is available at https://github.com/boxuLibrary/drt-vio-init.

We propose an approach based on convex relaxations for certifiably optimal robust multiview triangulation. To this end, we extend existing relaxation approaches to non-robust multiview triangulation by incorporating a least squares cost function. We propose two formulations, one based on epipolar constraints and one based on fractional reprojection constraints. The first is lower dimensional and remains tight under moderate noise and outlier levels, while the second is higher dimensional and therefore slower but remains tight even under extreme noise and outlier levels. We demonstrate through extensive experiments that the proposed approaches allow us to compute provably optimal reconstructions even under significant noise and a large percentage of outliers.

Recently, deep learning based approaches have shown promising results in 3D hand reconstruction from a single RGB image. These approaches can be roughly divided into model-based approaches, which are heavily dependent on the model’s parameter space, and model-free approaches, which require large numbers of 3D ground truths to reduce depth ambiguity and struggle in weakly-supervised scenarios. To overcome these issues, we propose a novel probabilistic model to achieve the robustness of model-based approaches and reduced dependence on the model’s parameter space of model-free approaches. The proposed probabilistic model incorporates a model-based network as a prior-net to estimate the prior probability distribution of joints and vertices. An Attention-based Mesh Vertices Uncertainty Regression (AMVUR) model is proposed to capture dependencies among vertices and the correlation between joints and mesh vertices to improve their feature representation. We further propose a learning based occlusion-aware Hand Texture Regression model to achieve high-fidelity texture reconstruction. We demonstrate the flexibility of the proposed probabilistic model to be trained in both supervised and weakly-supervised scenarios. The experimental results demonstrate our probabilistic model’s state-of-the-art accuracy in 3D hand and texture reconstruction from a single image in both training schemes, including in the presence of severe occlusions.

Existing methods for capturing datasets of 3D heads in dense semantic correspondence are slow and commonly address the problem in two separate steps; multi-view stereo (MVS) reconstruction followed by non-rigid registration. To simplify this process, we introduce TEMPEH (Towards Estimation of 3D Meshes from Performances of Expressive Heads) to directly infer 3D heads in dense correspondence from calibrated multi-view images. Registering datasets of 3D scans typically requires manual parameter tuning to find the right balance between accurately fitting the scans’ surfaces and being robust to scanning noise and outliers. Instead, we propose to jointly register a 3D head dataset while training TEMPEH. Specifically, during training, we minimize a geometric loss commonly used for surface registration, effectively leveraging TEMPEH as a regularizer. Our multi-view head inference builds on a volumetric feature representation that samples and fuses features from each view using camera calibration information. To account for partial occlusions and a large capture volume that enables head movements, we use view- and surface-aware feature fusion, and a spatial transformer-based head localization module, respectively. We use raw MVS scans as supervision during training, but, once trained, TEMPEH directly predicts 3D heads in dense correspondence without requiring scans. Predicting one head takes about …

Learning-based methods to solve dense 3D vision problems typically train on 3D sensor data. The respectively used principle of measuring distances provides advantages and drawbacks. These are typically not compared nor discussed in the literature due to a lack of multi-modal datasets. Texture-less regions are problematic for structure from motion and stereo, reflective material poses issues for active sensing, and distances for translucent objects are intricate to measure with existing hardware. Training on inaccurate or corrupt data induces model bias and hampers generalisation capabilities. These effects remain unnoticed if the sensor measurement is considered as ground truth during the evaluation. This paper investigates the effect of sensor errors for the dense 3D vision tasks of depth estimation and reconstruction. We rigorously show the significant impact of sensor characteristics on the learned predictions and notice generalisation issues arising from various technologies in everyday household environments. For evaluation, we introduce a carefully designed dataset comprising measurements from commodity sensors, namely D-ToF, I-ToF, passive/active stereo, and monocular RGB+P. Our study quantifies the considerable sensor noise impact and paves the way to improved dense vision estimates and targeted data fusion.

Holistic 3D scene understanding entails estimation of both layout configuration and object geometry in a 3D environment. Recent works have shown advances in 3D scene estimation from various input modalities (e.g., images, 3D scans), by leveraging 3D supervision (e.g., 3D bounding boxes or CAD models), for which collection at scale is expensive and often intractable. To address this shortcoming, we propose a new method to learn 3D scene priors of layout and shape without requiring any 3D ground truth. Instead, we rely on 2D supervision from multi-view RGB images. Our method represents a 3D scene as a latent vector, from which we can progressively decode to a sequence of objects characterized by their class categories, 3D bounding boxes, and meshes. With our trained autoregressive decoder representing the scene prior, our method facilitates many downstream applications, including scene synthesis, interpolation, and single-view reconstruction. Experiments on 3D-FRONT and ScanNet show that our method outperforms state of the art in single-view reconstruction, and achieves state-of-the-art results in scene synthesis against baselines which require for 3D supervision.

Recent advances in modeling 3D objects mostly rely on synthetic datasets due to the lack of large-scale real-scanned 3D databases. To facilitate the development of 3D perception, reconstruction, and generation in the real world, we propose OmniObject3D, a large vocabulary 3D object dataset with massive high-quality real-scanned 3D objects. OmniObject3D has several appealing properties: 1) Large Vocabulary: It comprises 6,000 scanned objects in 190 daily categories, sharing common classes with popular 2D datasets (e.g., ImageNet and LVIS), benefiting the pursuit of generalizable 3D representations. 2) Rich Annotations: Each 3D object is captured with both 2D and 3D sensors, providing textured meshes, point clouds, multiview rendered images, and multiple real-captured videos. 3) Realistic Scans: The professional scanners support high-quality object scans with precise shapes and realistic appearances. With the vast exploration space offered by OmniObject3D, we carefully set up four evaluation tracks: a) robust 3D perception, b) novel-view synthesis, c) neural surface reconstruction, and d) 3D object generation. Extensive studies are performed on these four benchmarks, revealing new observations, challenges, and opportunities for future research in realistic 3D vision.

Traditional 3D scene understanding approaches rely on labeled 3D datasets to train a model for a single task with supervision. We propose OpenScene, an alternative approach where a model predicts dense features for 3D scene points that are co-embedded with text and image pixels in CLIP feature space. This zero-shot approach enables task-agnostic training and open-vocabulary queries. For example, to perform SOTA zero-shot 3D semantic segmentation it first infers CLIP features for every 3D point and later classifies them based on similarities to embeddings of arbitrary class labels. More interestingly, it enables a suite of open-vocabulary scene understanding applications that have never been done before. For example, it allows a user to enter an arbitrary text query and then see a heat map indicating which parts of a scene match. Our approach is effective at identifying objects, materials, affordances, activities, and room types in complex 3D scenes, all using a single model trained without any labeled 3D data.

We present a method for 3D reconstruction only using calibrated multi-view surface azimuth maps. Our method, multi-view azimuth stereo, is effective for textureless or specular surfaces, which are difficult for conventional multi-view stereo methods. We introduce the concept of tangent space consistency: Multi-view azimuth observations of a surface point should be lifted to the same tangent space. Leveraging this consistency, we recover the shape by optimizing a neural implicit surface representation. Our method harnesses the robust azimuth estimation capabilities of photometric stereo methods or polarization imaging while bypassing potentially complex zenith angle estimation. Experiments using azimuth maps from various sources validate the accurate shape recovery with our method, even without zenith angles.

To effectively interrogate UAV-based images for detecting objects of interest, such as humans, it is essential to acquire large-scale UAV-based datasets that include human instances with various poses captured from widely varying viewing angles. As a viable alternative to laborious and costly data curation, we introduce Progressive Transformation Learning (PTL), which gradually augments a training dataset by adding transformed virtual images with enhanced realism. Generally, a virtual2real transformation generator in the conditional GAN framework suffers from quality degradation when a large domain gap exists between real and virtual images. To deal with the domain gap, PTL takes a novel approach that progressively iterates the following three steps: 1) select a subset from a pool of virtual images according to the domain gap, 2) transform the selected virtual images to enhance realism, and 3) add the transformed virtual images to the training set while removing them from the pool. In PTL, accurately quantifying the domain gap is critical. To do that, we theoretically demonstrate that the feature representation space of a given object detector can be modeled as a multivariate Gaussian distribution from which the Mahalanobis distance between a virtual object and the Gaussian distribution of each object category in the …

We address 2D floorplan reconstruction from 3D scans. Existing approaches typically employ heuristically designed multi-stage pipelines. Instead, we formulate floorplan reconstruction as a single-stage structured prediction task: find a variable-size set of polygons, which in turn are variable-length sequences of ordered vertices. To solve it we develop a novel Transformer architecture that generates polygons of multiple rooms in parallel, in a holistic manner without hand-crafted intermediate stages. The model features two-level queries for polygons and corners, and includes polygon matching to make the network end-to-end trainable. Our method achieves a new state-of-the-art for two challenging datasets, Structured3D and SceneCAD, along with significantly faster inference than previous methods. Moreover, it can readily be extended to predict additional information, i.e., semantic room types and architectural elements like doors and windows. Our code and models are available at: https://github.com/ywyue/RoomFormer.

We introduce a novel framework for training deep stereo networks effortlessly and without any ground-truth. By leveraging state-of-the-art neural rendering solutions, we generate stereo training data from image sequences collected with a single handheld camera. On top of them, a NeRF-supervised training procedure is carried out, from which we exploit rendered stereo triplets to compensate for occlusions and depth maps as proxy labels. This results in stereo networks capable of predicting sharp and detailed disparity maps. Experimental results show that models trained under this regime yield a 30-40% improvement over existing self-supervised methods on the challenging Middlebury dataset, filling the gap to supervised models and, most times, outperforming them at zero-shot generalization.

Semantic Scene Completion (SSC) transforms an image of single-view depth and/or RGB 2D pixels into 3D voxels, each of whose semantic labels are predicted. SSC is a well-known ill-posed problem as the prediction model has to “imagine” what is behind the visible surface, which is usually represented by Truncated Signed Distance Function (TSDF). Due to the sensory imperfection of the depth camera, most existing methods based on the noisy TSDF estimated from depth values suffer from 1) incomplete volumetric predictions and 2) confused semantic labels. To this end, we use the ground-truth 3D voxels to generate a perfect visible surface, called TSDF-CAD, and then train a “cleaner” SSC model. As the model is noise-free, it is expected to focus more on the “imagination” of unseen voxels. Then, we propose to distill the intermediate “cleaner” knowledge into another model with noisy TSDF input. In particular, we use the 3D occupancy feature and the semantic relations of the “cleaner self” to supervise the counterparts of the “noisy self” to respectively address the above two incorrect predictions. Experimental results validate that the proposed method improves the noisy counterparts with 3.1% IoU and 2.2% mIoU for measuring scene completion and SSC, and also achieves …

Indoor 360 panoramas have two essential properties. (1) The panoramas are continuous and seamless in the horizontal direction. (2) Gravity plays an important role in indoor environment design. By leveraging these properties, we present PanelNet, a framework that understands indoor environments using a novel panel representation of 360 images. We represent an equirectangular projection (ERP) as consecutive vertical panels with corresponding 3D panel geometry. To reduce the negative impact of panoramic distortion, we incorporate a panel geometry embedding network that encodes both the local and global geometric features of a panel. To capture the geometric context in room design, we introduce Local2Global Transformer, which aggregates local information within a panel and panel-wise global context. It greatly improves the model performance with low training overhead. Our method outperforms existing methods on indoor 360 depth estimation and shows competitive results against state-of-the-art approaches on the task of indoor layout estimation and semantic segmentation.

In this paper, we propose an approach for view-time interpolation of stereo videos. Specifically, we build upon X-Fields that approximates an interpolatable mapping between the input coordinates and 2D RGB images using a convolutional decoder. Our main contribution is to analyze and identify the sources of the problems with using X-Fields in our application and propose novel techniques to overcome these challenges. Specifically, we observe that X-Fields struggles to implicitly interpolate the disparities for large baseline cameras. Therefore, we propose multi-plane disparities to reduce the spatial distance of the objects in the stereo views. Moreover, we propose non-uniform time coordinates to handle the non-linear and sudden motion spikes in videos. We additionally introduce several simple, but important, improvements over X-Fields. We demonstrate that our approach is able to produce better results than the state of the art, while running in near real-time rates and having low memory and storage costs.

Depth estimation from indoor panoramas is challenging due to the equirectangular distortions of panoramas and inaccurate matching. In this paper, we propose a practical framework to improve the accuracy and efficiency of depth estimation from multi-view indoor panoramic images with the Neural Radiance Field technology. Specifically, we develop two networks to implicitly learn the Signed Distance Function for depth measurements and the radiance field from panoramas. We also introduce a novel spherical position embedding scheme to achieve high accuracy. For better convergence, we propose an initialization method for the network weights based on the Manhattan World Assumption. Furthermore, we devise a geometric consistency loss, leveraging the surface normal, to further refine the depth estimation. The experimental results demonstrate that our proposed method outperforms state-of-the-art works by a large margin in both quantitative and qualitative evaluations. Our source code is available at https://github.com/WJ-Chang-42/IndoorPanoDepth.

In recent years, there has been a significant increase in focus on the interpolation task of computer vision. Despite the tremendous advancement of video interpolation, point cloud interpolation remains insufficiently explored. Meanwhile, the existence of numerous nonlinear large motions in real-world scenarios makes the point cloud interpolation task more challenging. In light of these issues, we present NeuralPCI: an end-to-end 4D spatio-temporal Neural field for 3D Point Cloud Interpolation, which implicitly integrates multi-frame information to handle nonlinear large motions for both indoor and outdoor scenarios. Furthermore, we construct a new multi-frame point cloud interpolation dataset called NL-Drive for large nonlinear motions in autonomous driving scenes to better demonstrate the superiority of our method. Ultimately, NeuralPCI achieves state-of-the-art performance on both DHB (Dynamic Human Bodies) and NL-Drive datasets. Beyond the interpolation task, our method can be naturally extended to point cloud extrapolation, morphing, and auto-labeling, which indicates substantial potential in other domains. Codes are available at https://github.com/ispc-lab/NeuralPCI.

This paper presents a learning-based method for multi-view depth estimation from posed images. Our core idea is a “learning-to-optimize” paradigm that iteratively indexes a plane-sweeping cost volume and regresses the depth map via a convolutional Gated Recurrent Unit (GRU). Since the cost volume plays a paramount role in encoding the multi-view geometry, we aim to improve its construction both at pixel- and frame- levels. At the pixel level, we propose to break the symmetry of the Siamese network (which is typically used in MVS to extract image features) by introducing a transformer block to the reference image (but not to the source images). Such an asymmetric volume allows the network to extract global features from the reference image to predict its depth map. Given potential inaccuracies in the poses between reference and source images, we propose to incorporate a residual pose network to correct the relative poses. This essentially rectifies the cost volume at the frame level. We conduct extensive experiments on real-world MVS datasets and show that our method achieves state-of-the-art performance in terms of both within-dataset evaluation and cross-dataset generalization.

This paper presents an end-to-end neural mapping method for camera localization, dubbed NeuMap, encoding a whole scene into a grid of latent codes, with which a Transformer-based auto-decoder regresses 3D coordinates of query pixels. State-of-the-art feature matching methods require each scene to be stored as a 3D point cloud with per-point features, consuming several gigabytes of storage per scene. While compression is possible, performance drops significantly at high compression rates. Conversely, coordinate regression methods achieve high compression by storing scene information in a neural network but suffer from reduced robustness. NeuMap combines the advantages of both approaches by utilizing 1) learnable latent codes for efficient scene representation and 2) a scene-agnostic Transformer-based auto-decoder to infer coordinates for query pixels. This scene-agnostic network design learns robust matching priors from large-scale data and enables rapid optimization of codes for new scenes while keeping the network weights fixed. Extensive evaluations on five benchmarks show that NeuMap significantly outperforms other coordinate regression methods and achieves comparable performance to feature matching methods while requiring a much smaller scene representation size. For example, NeuMap achieves 39.1% accuracy in the Aachen night benchmark with only 6MB of data, whereas alternative methods require 100MB or several gigabytes and …

We introduce a method that simultaneously learns to explore new large environments and to reconstruct them in 3D from color images only. This is closely related to the Next Best View problem (NBV), where one has to identify where to move the camera next to improve the coverage of an unknown scene. However, most of the current NBV methods rely on depth sensors, need 3D supervision and/or do not scale to large scenes. Our method requires only a color camera and no 3D supervision. It simultaneously learns in a self-supervised fashion to predict a volume occupancy field from color images and, from this field, to predict the NBV. Thanks to this approach, our method performs well on new scenes as it is not biased towards any training 3D data. We demonstrate this on a recent dataset made of various 3D scenes and show it performs even better than recent methods requiring a depth sensor, which is not a realistic assumption for outdoor scenes captured with a flying drone.

We present vMAP, an object-level dense SLAM system using neural field representations. Each object is represented by a small MLP, enabling efficient, watertight object modelling without the need for 3D priors. As an RGB-D camera browses a scene with no prior information, vMAP detects object instances on-the-fly, and dynamically adds them to its map. Specifically, thanks to the power of vectorised training, vMAP can optimise as many as 50 individual objects in a single scene, with an extremely efficient training speed of 5Hz map update. We experimentally demonstrate significantly improved scene-level and object-level reconstruction quality compared to prior neural field SLAM systems. Project page: https://kxhit.github.io/vMAP.

What is a rose, visually? A rose comprises its intrinsics, including the distribution of geometry, texture, and material specific to its object category. With knowledge of these intrinsic properties, we may render roses of different sizes and shapes, in different poses, and under different lighting conditions. In this work, we build a generative model that learns to capture such object intrinsics from a single image, such as a photo of a bouquet. Such an image includes multiple instances of an object type. These instances all share the same intrinsics, but appear different due to a combination of variance within these intrinsics and differences in extrinsic factors, such as pose and illumination. Experiments show that our model successfully learns object intrinsics (distribution of geometry, texture, and material) for a wide range of objects, each from a single Internet image. Our method achieves superior results on multiple downstream tasks, including intrinsic image decomposition, shape and image generation, view synthesis, and relighting.

Non-line-of-sight (NLOS) tracking has drawn increasing attention in recent years, due to its ability to detect object motion out of sight. Most previous works on NLOS tracking rely on active illumination, e.g., laser, and suffer from high cost and elaborate experimental conditions. Besides, these techniques are still far from practical application due to oversimplified settings. In contrast, we propose a purely passive method to track a person walking in an invisible room by only observing a relay wall, which is more in line with real application scenarios, e.g., security. To excavate imperceptible changes in videos of the relay wall, we introduce difference frames as an essential carrier of temporal-local motion messages. In addition, we propose PAC-Net, which consists of alternating propagation and calibration, making it capable of leveraging both dynamic and static messages on a frame-level granularity. To evaluate the proposed method, we build and publish the first dynamic passive NLOS tracking dataset, NLOS-Track, which fills the vacuum of realistic NLOS datasets. NLOS-Track contains thousands of NLOS video clips and corresponding trajectories. Both real-shot and synthetic data are included. Our codes and dataset are available at https://againstentropy.github.io/NLOS-Track/.

Existing autonomous vehicles primarily use sensors that rely on electromagnetic waves which are undisturbed in good environmental conditions but can suffer in adverse scenarios, such as low light or for objects with low reflectance. Moreover, only objects in direct line-of-sight are typically detected by these existing methods. Acoustic pressure waves emanating from road users do not share these limitations. However, such signals are typically ignored in automotive perception because they suffer from low spatial resolution and lack directional information. In this work, we introduce long-range acoustic beamforming of pressure waves from noise directly produced by automotive vehicles in-the-wild as a {complementary sensing modality} to traditional optical sensor approaches for detection of objects in dynamic traffic environments. To this end, we introduce the first multimodal long-range acoustic beamforming dataset. We propose a neural aperture expansion method for beamforming and we validate its utility for multimodal automotive object detection. We validate the benefit of adding sound detections to existing RGB cameras in challenging automotive scenarios, where camera-only approaches fail or do not deliver the ultra-fast rates of pressure sensors.

Multi-camera 3D object detection blossoms in recent years and most of state-of-the-art methods are built up on the bird’s-eye-view (BEV) representations. Albeit remarkable performance, these works suffer from low efficiency. Typically, knowledge distillation can be used for model compression. However, due to unclear 3D geometry reasoning, expert features usually contain some noisy and confusing areas. In this work, we investigate on how to distill the knowledge from an imperfect expert. We propose FD3D, a Focal Distiller for 3D object detection. Specifically, a set of queries are leveraged to locate the instance-level areas for masked feature generation, to intensify feature representation ability in these areas. Moreover, these queries search out the representative fine-grained positions for refined distillation. We verify the effectiveness of our method by applying it to two popular detection models, BEVFormer and DETR3D. The results demonstrate that our method achieves improvements of 4.07 and 3.17 points respectively in terms of NDS metric on nuScenes benchmark. Code is hosted at https://github.com/OpenPerceptionX/BEVPerception-Survey-Recipe.

3D lane detection which plays a crucial role in vehicle routing, has recently been a rapidly developing topic in autonomous driving. Previous works struggle with practicality due to their complicated spatial transformations and inflexible representations of 3D lanes. Faced with the issues, our work proposes an efficient and robust monocular 3D lane detection called BEV-LaneDet with three main contributions. First, we introduce the Virtual Camera that unifies the in/extrinsic parameters of cameras mounted on different vehicles to guarantee the consistency of the spatial relationship among cameras. It can effectively promote the learning procedure due to the unified visual space. We secondly propose a simple but efficient 3D lane representation called Key-Points Representation. This module is more suitable to represent the complicated and diverse 3D lane structures. At last, we present a light-weight and chip-friendly spatial transformation module named Spatial Transformation Pyramid to transform multiscale front-view features into BEV features. Experimental results demonstrate that our work outperforms the state-of-the-art approaches in terms of F-Score, being 10.6% higher on the OpenLane dataset and 4.0% higher on the Apollo 3D synthetic dataset, with a speed of 185 FPS. Code is released at https://github.com/gigo-team/bevlanedet.

3D object detection techniques commonly follow a pipeline that aggregates predicted object central point features to compute candidate points. However, these candidate points contain only positional information, largely ignoring the object-level shape information. This eventually leads to sub-optimal 3D object detection. In this work, we propose AShapeFormer, a semantics-guided object-level shape encoding module for 3D object detection. This is a plug-n-play module that leverages multi-head attention to encode object shape information. We also propose shape tokens and object-scene positional encoding to ensure that the shape information is fully exploited. Moreover, we introduce a semantic guidance sub-module to sample more foreground points and suppress the influence of background points for a better object shape perception. We demonstrate a straightforward enhancement of multiple existing methods with our AShapeFormer. Through extensive experiments on the popular SUN RGB-D and ScanNetV2 dataset, we show that our enhanced models are able to outperform the baselines by a considerable absolute margin of up to 8.1%. Code will be available at https://github.com/ZechuanLi/AShapeFormer

3D object detection is an important task in autonomous driving to perceive the surroundings. Despite the excellent performance, the existing 3D detectors lack the robustness to real-world corruptions caused by adverse weathers, sensor noises, etc., provoking concerns about the safety and reliability of autonomous driving systems. To comprehensively and rigorously benchmark the corruption robustness of 3D detectors, in this paper we design 27 types of common corruptions for both LiDAR and camera inputs considering real-world driving scenarios. By synthesizing these corruptions on public datasets, we establish three corruption robustness benchmarks---KITTI-C, nuScenes-C, and Waymo-C. Then, we conduct large-scale experiments on 24 diverse 3D object detection models to evaluate their corruption robustness. Based on the evaluation results, we draw several important findings, including: 1) motion-level corruptions are the most threatening ones that lead to significant performance drop of all models; 2) LiDAR-camera fusion models demonstrate better robustness; 3) camera-only models are extremely vulnerable to image corruptions, showing the indispensability of LiDAR point clouds. We release the benchmarks and codes at https://github.com/thu-ml/3DCorruptionsAD to be helpful for future studies.

Spherical image object detection emerges in many applications from virtual reality to robotics and automatic driving, while many existing detectors use ln-norms loss for regression of spherical bounding boxes. There are two intrinsic flaws for ln-norms loss, i.e., independent optimization of parameters and inconsistency between metric (dominated by IoU) and loss. These problems are common in planar image detection but more significant in spherical image detection. Solution for these problems has been extensively discussed in planar image detection by using IoU loss and related variants. However, these solutions cannot be migrated to spherical image object detection due to the undifferentiable of the Spherical IoU (SphIoU). In this paper, we design a simple but effective regression loss based on Gaussian Label Distribution Learning (GLDL) for spherical image object detection. Besides, we observe that the scale of the object in a spherical image varies greatly. The huge differences among objects from different categories make the sample selection strategy based on SphIoU challenging. Therefore, we propose GLDL-ATSS as a better training sample selection strategy for objects of the spherical image, which can alleviate the drawback of IoU threshold-based strategy of scale-sample imbalance. Extensive results on various two datasets with different baseline detectors show …

Robust and accurate geometric understanding against adverse weather conditions is one top prioritized conditions to achieve a high-level autonomy of self-driving cars. However, autonomous driving algorithms relying on the visible spectrum band are easily impacted by weather and lighting conditions. A long-wave infrared camera, also known as a thermal imaging camera, is a potential rescue to achieve high-level robustness. However, the missing necessities are the well-established large-scale dataset and public benchmark results. To this end, in this paper, we first built a large-scale Multi-Spectral Stereo (MS^2) dataset, including stereo RGB, stereo NIR, stereo thermal, and stereo LiDAR data along with GNSS/IMU information. The collected dataset provides about 195K synchronized data pairs taken from city, residential, road, campus, and suburban areas in the morning, daytime, and nighttime under clear-sky, cloudy, and rainy conditions. Secondly, we conduct an exhaustive validation process of monocular and stereo depth estimation algorithms designed on visible spectrum bands to benchmark their performance in the thermal image domain. Lastly, we propose a unified depth network that effectively bridges monocular depth and stereo depth tasks from a conditional random field approach perspective. Our dataset and source code are available at https://github.com/UkcheolShin/MS2-MultiSpectralStereoDataset.

Video-based gait recognition has achieved impressive results in constrained scenarios. However, visual cameras neglect human 3D structure information, which limits the feasibility of gait recognition in the 3D wild world. Instead of extracting gait features from images, this work explores precise 3D gait features from point clouds and proposes a simple yet efficient 3D gait recognition framework, termed LidarGait. Our proposed approach projects sparse point clouds into depth maps to learn the representations with 3D geometry information, which outperforms existing point-wise and camera-based methods by a significant margin. Due to the lack of point cloud datasets, we build the first large-scale LiDAR-based gait recognition dataset, SUSTech1K, collected by a LiDAR sensor and an RGB camera. The dataset contains 25,239 sequences from 1,050 subjects and covers many variations, including visibility, views, occlusions, clothing, carrying, and scenes. Extensive experiments show that (1) 3D structure information serves as a significant feature for gait recognition. (2) LidarGait outperforms existing point-based and silhouette-based methods by a significant margin, while it also offers stable cross-view results. (3) The LiDAR sensor is superior to the RGB camera for gait recognition in the outdoor environment. The source code and dataset have been made available at https://lidargait.github.io.

When deploying the Unmanned Aerial Vehicles object detection (UAV-OD) network to complex and unseen real-world scenarios, the generalization ability is usually reduced due to the domain shift. To address this issue, this paper proposes a novel frequency domain disentanglement method to improve the UAV-OD generalization. Specifically, we first verified that the spectrum of different bands in the image has different effects to the UAV-OD generalization. Based on this conclusion, we design two learnable filters to extract domain-invariant spectrum and domain-specific spectrum, respectively. The former can be used to train the UAV-OD network and improve its capacity for generalization. In addition, we design a new instance-level contrastive loss to guide the network training. This loss enables the network to concentrate on extracting domain-invariant spectrum and domain-specific spectrum, so as to achieve better disentangling results. Experimental results on three unseen target domains demonstrate that our method has better generalization ability than both the baseline method and state-of-the-art methods.

LiDAR provides accurate geometric measurements of the 3D world. Unfortunately, dense LiDARs are very expensive and the point clouds captured by low-beam LiDAR are often sparse. To address these issues, we present UltraLiDAR, a data-driven framework for scene-level LiDAR completion, LiDAR generation, and LiDAR manipulation. The crux of UltraLiDAR is a compact, discrete representation that encodes the point cloud’s geometric structure, is robust to noise, and is easy to manipulate. We show that by aligning the representation of a sparse point cloud to that of a dense point cloud, we can densify the sparse point clouds as if they were captured by a real high-density LiDAR, drastically reducing the cost. Furthermore, by learning a prior over the discrete codebook, we can generate diverse, realistic LiDAR point clouds for self-driving. We evaluate the effectiveness of UltraLiDAR on sparse-to-dense LiDAR completion and LiDAR generation. Experiments show that densifying real-world point clouds with our approach can significantly improve the performance of downstream perception systems. Compared to prior art on LiDAR generation, our approach generates much more realistic point clouds. According to A/B test, over 98.5% of the time human participants prefer our results over those of previous methods. Please refer to project page …

3D single object tracking plays an essential role in many applications, such as autonomous driving. It remains a challenging problem due to the large appearance variation and the sparsity of points caused by occlusion and limited sensor capabilities. Therefore, contextual information across two consecutive frames is crucial for effective object tracking. However, points containing such useful information are often overlooked and cropped out in existing methods, leading to insufficient use of important contextual knowledge. To address this issue, we propose CXTrack, a novel transformer-based network for 3D object tracking, which exploits ConteXtual information to improve the tracking results. Specifically, we design a target-centric transformer network that directly takes point features from two consecutive frames and the previous bounding box as input to explore contextual information and implicitly propagate target cues. To achieve accurate localization for objects of all sizes, we propose a transformer-based localization head with a novel center embedding module to distinguish the target from distractors. Extensive experiments on three large-scale datasets, KITTI, nuScenes and Waymo Open Dataset, show that CXTrack achieves state-of-the-art tracking performance while running at 34 FPS.

Robust and reliable semantic segmentation in complex scenes is crucial for many real-life applications such as autonomous safe driving and nighttime rescue. In most approaches, it is typical to make use of RGB images as input. They however work well only in preferred weather conditions; when facing adverse conditions such as rainy, overexposure, or low-light, they often fail to deliver satisfactory results. This has led to the recent investigation into multispectral semantic segmentation, where RGB and thermal infrared (RGBT) images are both utilized as input. This gives rise to significantly more robust segmentation of image objects in complex scenes and under adverse conditions. Nevertheless, the present focus in single RGBT image input restricts existing methods from well addressing dynamic real-world scenes. Motivated by the above observations, in this paper, we set out to address a relatively new task of semantic segmentation of multispectral video input, which we refer to as Multispectral Video Semantic Segmentation, or MVSS in short. An in-house MVSeg dataset is thus curated, consisting of 738 calibrated RGB and thermal videos, accompanied by 3,545 fine-grained pixel-level semantic annotations of 26 categories. Our dataset contains a wide range of challenging urban scenes in both daytime and nighttime. Moreover, we …

Extending the success of 2D Large Kernel to 3D perception is challenging due to: 1. the cubically-increasing overhead in processing 3D data; 2. the optimization difficulties from data scarcity and sparsity. Previous work has taken the first step to scale up the kernel size from 3x3x3 to 7x7x7 by introducing block-shared weights. However, to reduce the feature variations within a block, it only employs modest block size and fails to achieve larger kernels like the 21x21x21. To address this issue, we propose a new method, called LinK, to achieve a wider-range perception receptive field in a convolution-like manner with two core designs. The first is to replace the static kernel matrix with a linear kernel generator, which adaptively provides weights only for non-empty voxels. The second is to reuse the pre-computed aggregation results in the overlapped blocks to reduce computation complexity. The proposed method successfully enables each voxel to perceive context within a range of 21x21x21. Extensive experiments on two basic perception tasks, 3D object detection and 3D semantic segmentation, demonstrate the effectiveness of our method. Notably, we rank 1st on the public leaderboard of the 3D detection benchmark of nuScenes (LiDAR track), by simply incorporating a LinK-based backbone into …

Predicting how the world can evolve in the future is crucial for motion planning in autonomous systems. Classical methods are limited because they rely on costly human annotations in the form of semantic class labels, bounding boxes, and tracks or HD maps of cities to plan their motion -- and thus are difficult to scale to large unlabeled datasets. One promising self-supervised task is 3D point cloud forecasting from unannotated LiDAR sequences. We show that this task requires algorithms to implicitly capture (1) sensor extrinsics (i.e., the egomotion of the autonomous vehicle), (2) sensor intrinsics (i.e., the sampling pattern specific to the particular LiDAR sensor), and (3) the shape and motion of other objects in the scene. But autonomous systems should make predictions about the world and not their sensors! To this end, we factor out (1) and (2) by recasting the task as one of spacetime (4D) occupancy forecasting. But because it is expensive to obtain ground-truth 4D occupancy, we “render” point cloud data from 4D occupancy predictions given sensor extrinsics and intrinsics, allowing one to train and test occupancy algorithms with unannotated LiDAR sequences. This also allows one to evaluate and compare point cloud forecasting algorithms across diverse …

This paper explores the potential of curriculum learning in LiDAR-based 3D object detection by proposing a curricular object manipulation (COM) framework. The framework embeds the curricular training strategy into both the loss design and the augmentation process. For the loss design, we propose the COMLoss to dynamically predict object-level difficulties and emphasize objects of different difficulties based on training stages. On top of the widely-used augmentation technique called GT-Aug in LiDAR detection tasks, we propose a novel COMAug strategy which first clusters objects in ground-truth database based on well-designed heuristics. Group-level difficulties rather than individual ones are then predicted and updated during training for stable results. Model performance and generalization capabilities can be improved by sampling and augmenting progressively more difficult objects into the training points. Extensive experiments and ablation studies reveal the superior and generality of the proposed framework. The code is available at https://github.com/ZZY816/COM.

Multimodal fusion can make semantic segmentation more robust. However, fusing an arbitrary number of modalities remains underexplored. To delve into this problem, we create the DeLiVER arbitrary-modal segmentation benchmark, covering Depth, LiDAR, multiple Views, Events, and RGB. Aside from this, we provide this dataset in four severe weather conditions as well as five sensor failure cases to exploit modal complementarity and resolve partial outages. To facilitate this data, we present the arbitrary cross-modal segmentation model CMNeXt. It encompasses a Self-Query Hub (SQ-Hub) designed to extract effective information from any modality for subsequent fusion with the RGB representation and adds only negligible amounts of parameters (~0.01M) per additional modality. On top, to efficiently and flexibly harvest discriminative cues from the auxiliary modalities, we introduce the simple Parallel Pooling Mixer (PPX). With extensive experiments on a total of six benchmarks, our CMNeXt achieves state-of-the-art performance, allowing to scale from 1 to 81 modalities on the DeLiVER, KITTI-360, MFNet, NYU Depth V2, UrbanLF, and MCubeS datasets. On the freshly collected DeLiVER, the quad-modal CMNeXt reaches up to 66.30% in mIoU with a +9.10% gain as compared to the mono-modal baseline.

Learning-based outlier (mismatched correspondence) rejection for robust 3D registration generally formulates the outlier removal as an inlier/outlier classification problem. The core for this to be successful is to learn the discriminative inlier/outlier feature representations. In this paper, we develop a novel variational non-local network-based outlier rejection framework for robust alignment. By reformulating the non-local feature learning with variational Bayesian inference, the Bayesian-driven long-range dependencies can be modeled to aggregate discriminative geometric context information for inlier/outlier distinction. Specifically, to achieve such Bayesian-driven contextual dependencies, each query/key/value component in our non-local network predicts a prior feature distribution and a posterior one. Embedded with the inlier/outlier label, the posterior feature distribution is label-dependent and discriminative. Thus, pushing the prior to be close to the discriminative posterior in the training step enables the features sampled from this prior at test time to model high-quality long-range dependencies. Notably, to achieve effective posterior feature guidance, a specific probabilistic graphical model is designed over our non-local model, which lets us derive a variational low bound as our optimization objective for model training. Finally, we propose a voting-based inlier searching strategy to cluster the high-quality hypothetical inliers for transformation estimation. Extensive experiments on 3DMatch, 3DLoMatch, and KITTI datasets …

Training a 3D human keypoint detector from point clouds in a supervised manner requires large volumes of high quality labels. While it is relatively easy to capture large amounts of human point clouds, annotating 3D keypoints is expensive, subjective, error prone and especially difficult for long-tail cases (pedestrians with rare poses, scooterists, etc.). In this work, we propose GC-KPL - Geometry Consistency inspired Key Point Leaning, an approach for learning 3D human joint locations from point clouds without human labels. We achieve this by our novel unsupervised loss formulations that account for the structure and movement of the human body. We show that by training on a large training set from Waymo Open Dataset without any human annotated keypoints, we are able to achieve reasonable performance as compared to the fully supervised approach. Further, the backbone benefits from the unsupervised training and is useful in downstream fewshot learning of keypoints, where fine-tuning on only 10 percent of the labeled training data gives comparable performance to fine-tuning on the entire set. We demonstrated that GC-KPL outperforms by a large margin over SoTA when trained on entire dataset and efficiently leverages large volumes of unlabeled data.

Masked signal modeling has greatly advanced self-supervised pre-training for language and 2D images. However, it is still not fully explored in 3D scene understanding. Thus, this paper introduces Masked Shape Prediction (MSP), a new framework to conduct masked signal modeling in 3D scenes. MSP uses the essential 3D semantic cue, i.e., geometric shape, as the prediction target for masked points. The context-enhanced shape target consisting of explicit shape context and implicit deep shape feature is proposed to facilitate exploiting contextual cues in shape prediction. Meanwhile, the pre-training architecture in MSP is carefully designed to alleviate the masked shape leakage from point coordinates. Experiments on multiple 3D understanding tasks on both indoor and outdoor datasets demonstrate the effectiveness of MSP in learning good feature representations to consistently boost downstream performance.

The recognition capabilities of current state-of-the-art 3D models are limited by datasets with a small number of annotated data and a pre-defined set of categories. In its 2D counterpart, recent advances have shown that similar problems can be significantly alleviated by employing knowledge from other modalities, such as language. Inspired by this, leveraging multimodal information for 3D modality could be promising to improve 3D understanding under the restricted data regime, but this line of research is not well studied. Therefore, we introduce ULIP to learn a unified representation of images, language, and 3D point clouds by pre-training with object triplets from the three modalities. To overcome the shortage of training triplets, ULIP leverages a pre-trained vision-language model that has already learned a common visual and textual space by training with massive image-text pairs. Then, ULIP learns a 3D representation space aligned with the common image-text space, using a small number of automatically synthesized triplets. ULIP is agnostic to 3D backbone networks and can easily be integrated into any 3D architecture. Experiments show that ULIP effectively improves the performance of multiple recent 3D backbones by simply pre-training them on ShapeNet55 using our framework, achieving state-of-the-art performance in both standard 3D classification …

The goal of open-vocabulary detection is to identify novel objects based on arbitrary textual descriptions. In this paper, we address open-vocabulary 3D point-cloud detection by a dividing-and-conquering strategy, which involves: 1) developing a point-cloud detector that can learn a general representation for localizing various objects, and 2) connecting textual and point-cloud representations to enable the detector to classify novel object categories based on text prompting. Specifically, we resort to rich image pre-trained models, by which the point-cloud detector learns localizing objects under the supervision of predicted 2D bounding boxes from 2D pre-trained detectors. Moreover, we propose a novel de-biased triplet cross-modal contrastive learning to connect the modalities of image, point-cloud and text, thereby enabling the point-cloud detector to benefit from vision-language pre-trained models, i.e., CLIP. The novel use of image and vision-language pre-trained models for point-cloud detectors allows for open-vocabulary 3D object detection without the need for 3D annotations. Experiments demonstrate that the proposed method improves at least 3.03 points and 7.47 points over a wide range of baselines on the ScanNet and SUN RGB-D datasets, respectively. Furthermore, we provide a comprehensive analysis to explain why our approach works.

Transformer, as an alternative to CNN, has been proven effective in many modalities (e.g., texts and images). For 3D point cloud transformers, existing efforts focus primarily on pushing their accuracy to the state-of-the-art level. However, their latency lags behind sparse convolution-based models (3x slower), hindering their usage in resource-constrained, latency-sensitive applications (such as autonomous driving). This inefficiency comes from point clouds’ sparse and irregular nature, whereas transformers are designed for dense, regular workloads. This paper presents FlatFormer to close this latency gap by trading spatial proximity for better computational regularity. We first flatten the point cloud with window-based sorting and partition points into groups of equal sizes rather than windows of equal shapes. This effectively avoids expensive structuring and padding overheads. We then apply self-attention within groups to extract local features, alternate sorting axis to gather features from different directions, and shift windows to exchange features across groups. FlatFormer delivers state-of-the-art accuracy on Waymo Open Dataset with 4.6x speedup over (transformer-based) SST and 1.4x speedup over (sparse convolutional) CenterPoint. This is the first point cloud transformer that achieves real-time performance on edge GPUs and is faster than sparse convolutional methods while achieving on-par or even superior accuracy on large-scale benchmarks.

Self-supervised learning can extract representations of good quality from solely unlabeled data, which is appealing for point cloud videos due to their high labelling cost. In this paper, we propose a contrastive mask prediction (PointCMP) framework for self-supervised learning on point cloud videos. Specifically, our PointCMP employs a two-branch structure to achieve simultaneous learning of both local and global spatio-temporal information. On top of this two-branch structure, a mutual similarity based augmentation module is developed to synthesize hard samples at the feature level. By masking dominant tokens and erasing principal channels, we generate hard samples to facilitate learning representations with better discrimination and generalization performance. Extensive experiments show that our PointCMP achieves the state-of-the-art performance on benchmark datasets and outperforms existing full-supervised counterparts. Transfer learning results demonstrate the superiority of the learned representations across different datasets and tasks.

This paper proposes a convolution structure for learning SE(3)-equivariant features from 3D point clouds. It can be viewed as an equivariant version of kernel point convolutions (KPConv), a widely used convolution form to process point cloud data. Compared with existing equivariant networks, our design is simple, lightweight, fast, and easy to be integrated with existing task-specific point cloud learning pipelines. We achieve these desirable properties by combining group convolutions and quotient representations. Specifically, we discretize SO(3) to finite groups for their simplicity while using SO(2) as the stabilizer subgroup to form spherical quotient feature fields to save computations. We also propose a permutation layer to recover SO(3) features from spherical features to preserve the capacity to distinguish rotations. Experiments show that our method achieves comparable or superior performance in various tasks, including object classification, pose estimation, and keypoint-matching, while consuming much less memory and running faster than existing work. The proposed method can foster the development of equivariant models for real-world applications based on point clouds.

In this work, we show that it is feasible to perform multiple tasks concurrently on point cloud with a straightforward yet effective multi-task network. Our framework, Poly-PC, tackles the inherent obstacles (e.g., different model architectures caused by task bias and conflicting gradients caused by multiple dataset domains, etc.) of multi-task learning on point cloud. Specifically, we propose a residual set abstraction (Res-SA) layer for efficient and effective scaling in both width and depth of the network, hence accommodating the needs of various tasks. We develop a weight-entanglement-based one-shot NAS technique to find optimal architectures for all tasks. Moreover, such technique entangles the weights of multiple tasks in each layer to offer task-shared parameters for efficient storage deployment while providing ancillary task-specific parameters for learning task-related features. Finally, to facilitate the training of Poly-PC, we introduce a task-prioritization-based gradient balance algorithm that leverages task prioritization to reconcile conflicting gradients, ensuring high performance for all tasks. Benefiting from the suggested techniques, models optimized by Poly-PC collectively for all tasks keep fewer total FLOPs and parameters and outperform previous methods. We also demonstrate that Poly-PC allows incremental learning and evades catastrophic forgetting when tuned to a new task.

Recently, self-attention networks achieve impressive performance in point cloud segmentation due to their superiority in modeling long-range dependencies. However, compared to self-attention mechanism, we find graph convolutions show a stronger ability in capturing local geometry information with less computational cost. In this paper, we employ a hybrid architecture design to construct our Graph Convolution Network with Attentive Filtering (AF-GCN), which takes advantage of both graph convolution and self-attention mechanism. We adopt graph convolutions to aggregate local features in the shallow encoder stages, while in the deeper stages, we propose a self-attention-like module named Graph Attentive Filter (GAF) to better model long-range contexts from distant neighbors. Besides, to further improve graph representation for point cloud segmentation, we employ a Spatial Feature Projection (SFP) module for graph convolutions which helps to handle spatial variations of unstructured point clouds. Finally, a graph-shared down-sampling and up-sampling strategy is introduced to make full use of the graph structures in point cloud processing. We conduct extensive experiments on multiple datasets including S3DIS, ScanNetV2, Toronto-3D, and ShapeNetPart. Experimental results show our AF-GCN obtains competitive performance.

An ideal point cloud registration framework should have superior accuracy, acceptable efficiency, and strong generalizability. However, this is highly challenging since existing registration techniques are either not accurate enough, far from efficient, or generalized poorly. It remains an open question that how to achieve a satisfying balance between this three key elements. In this paper, we propose BUFFER, a point cloud registration method for balancing accuracy, efficiency, and generalizability. The key to our approach is to take advantage of both point-wise and patch-wise techniques, while overcoming the inherent drawbacks simultaneously. Different from a simple combination of existing methods, each component of our network has been carefully crafted to tackle specific issues. Specifically, a Point-wise Learner is first introduced to enhance computational efficiency by predicting keypoints and improving the representation capacity of features by estimating point orientations, a Patch-wise Embedder which leverages a lightweight local feature learner is then deployed to extract efficient and general patch features. Additionally, an Inliers Generator which combines simple neural layers and general features is presented to search inlier correspondences. Extensive experiments on real-world scenarios demonstrate that our method achieves the best of both worlds in accuracy, efficiency, and generalization. In particular, our method not only …

Rapid development in automatic vector extraction from remote sensing images has been witnessed in recent years. However, the vast majority of existing works concentrate on a specific target, fragile to category variety, and hardly achieve stable performance crossing different categories. In this work, we propose an innovative class-agnostic model, namely TopDiG, to directly extract topological directional graphs from remote sensing images and solve these issues. Firstly, TopDiG employs a topology-concentrated node detector (TCND) to detect nodes and obtain compact perception of topological components. Secondly, we propose a dynamic graph supervision (DGS) strategy to dynamically generate adjacency graph labels from unordered nodes. Finally, the directional graph (DiG) generator module is designed to construct topological directional graphs from predicted nodes. Experiments on the Inria, CrowdAI, GID, GF2 and Massachusetts datasets empirically demonstrate that TopDiG is class-agnostic and achieves competitive performance on all datasets.

Rigid structures such as cars or any other solid objects are often represented by finite clouds of unlabeled points. The most natural equivalence on these point clouds is rigid motion or isometry maintaining all inter-point distances. Rigid patterns of point clouds can be reliably compared only by complete isometry invariants that can also be called equivariant descriptors without false negatives (isometric clouds having different descriptions) and without false positives (non-isometric clouds with the same description). Noise and motion in data motivate a search for invariants that are continuous under perturbations of points in a suitable metric. We propose the first continuous and complete invariant of unlabeled clouds in any Euclidean space. For a fixed dimension, the new metric for this invariant is computable in a polynomial time in the number of points.

The ability of scene understanding has sparked active research for panoramic image semantic segmentation. However, the performance is hampered by distortion of the equirectangular projection (ERP) and a lack of pixel-wise annotations. For this reason, some works treat the ERP and pinhole images equally and transfer knowledge from the pinhole to ERP images via unsupervised domain adaptation (UDA). However, they fail to handle the domain gaps caused by: 1) the inherent differences between camera sensors and captured scenes; 2) the distinct image formats (e.g., ERP and pinhole images). In this paper, we propose a novel yet flexible dual-path UDA framework, DPPASS, taking ERP and tangent projection (TP) images as inputs. To reduce the domain gaps, we propose cross-projection and intra-projection training. The cross-projection training includes tangent-wise feature contrastive training and prediction consistency training. That is, the former formulates the features with the same projection locations as positive examples and vice versa, for the models’ awareness of distortion, while the latter ensures the consistency of cross-model predictions between the ERP and TP. Moreover, adversarial intra-projection training is proposed to reduce the inherent gap, between the features of the pinhole images and those of the ERP and TP images, respectively. Importantly, the …

Jointly matching multiple, non-rigidly deformed 3D shapes is a challenging, NP-hard problem. A perfect matching is necessarily cycle-consistent: Following the pairwise point correspondences along several shapes must end up at the starting vertex of the original shape. Unfortunately, existing quantum shape-matching methods do not support multiple shapes and even less cycle consistency. This paper addresses the open challenges and introduces the first quantum-hybrid approach for 3D shape multi-matching; in addition, it is also cycle-consistent. Its iterative formulation is admissible to modern adiabatic quantum hardware and scales linearly with the total number of input shapes. Both these characteristics are achieved by reducing the N-shape case to a sequence of three-shape matchings, the derivation of which is our main technical contribution. Thanks to quantum annealing, high-quality solutions with low energy are retrieved for the intermediate NP-hard objectives. On benchmark datasets, the proposed approach significantly outperforms extensions to multi-shape matching of a previous quantum-hybrid two-shape matching method and is on-par with classical multi-matching methods. Our source code is available at 4dqv.mpi-inf.mpg.de/CCuantuMM/

Local feature extraction is a standard approach in computer vision for tackling important tasks such as image matching and retrieval. The core assumption of most methods is that images undergo affine transformations, disregarding more complicated effects such as non-rigid deformations. Furthermore, incipient works tailored for non-rigid correspondence still rely on keypoint detectors designed for rigid transformations, hindering performance due to the limitations of the detector. We propose DALF (Deformation-Aware Local Features), a novel deformation-aware network for jointly detecting and describing keypoints, to handle the challenging problem of matching deformable surfaces. All network components work cooperatively through a feature fusion approach that enforces the descriptors’ distinctiveness and invariance. Experiments using real deforming objects showcase the superiority of our method, where it delivers 8% improvement in matching scores compared to the previous best results. Our approach also enhances the performance of two real-world applications: deformable object retrieval and non-rigid 3D surface registration. Code for training, inference, and applications are publicly available at verlab.dcc.ufmg.br/descriptors/dalf_cvpr23.

Deep functional maps have recently emerged as a successful paradigm for non-rigid 3D shape correspondence tasks. An essential step in this pipeline consists in learning feature functions that are used as constraints to solve for a functional map inside the network. However, the precise nature of the information learned and stored in these functions is not yet well understood. Specifically, a major question is whether these features can be used for any other objective, apart from their purely algebraic role, in solving for functional map matrices. In this paper, we show that under some mild conditions, the features learned within deep functional map approaches can be used as point-wise descriptors and thus are directly comparable across different shapes, even without the necessity of solving for a functional map at test time. Furthermore, informed by our analysis, we propose effective modifications to the standard deep functional map pipeline, which promotes structural properties of learned features, significantly improving the matching results. Finally, we demonstrate that previously unsuccessful attempts at using extrinsic architectures for deep functional map feature extraction can be remedied via simple architectural changes, which promote the theoretical properties suggested by our analysis. We thus bridge the gap between intrinsic and …

In the field of binocular stereo matching, remarkable progress has been made by iterative methods like RAFT-Stereo and CREStereo. However, most of these methods lose information during the iterative process, making it difficult to generate more detailed difference maps that take full advantage of high-frequency information. We propose the Decouple module to alleviate the problem of data coupling and allow features containing subtle details to transfer across the iterations which proves to alleviate the problem significantly in the ablations. To further capture high-frequency details, we propose a Normalization Refinement module that unifies the disparities as a proportion of the disparities over the width of the image, which address the problem of module failure in cross-domain scenarios. Further, with the above improvements, the ResNet-like feature extractor that has not been changed for years becomes a bottleneck. Towards this end, we proposed a multi-scale and multi-stage feature extractor that introduces the channel-wise self-attention mechanism which greatly addresses this bottleneck. Our method (DLNR) ranks 1st on the Middlebury leaderboard, significantly outperforming the next best method by 13.04%. Our method also achieves SOTA performance on the KITTI-2015 benchmark for D1-fg.

Optical flow estimation is a challenging problem remaining unsolved. Recent deep learning based optical flow models have achieved considerable success. However, these models often train networks from the scratch on standard optical flow data, which restricts their ability to robustly and geometrically match image features. In this paper, we propose a rethinking to previous optical flow estimation. We particularly leverage Geometric Image Matching (GIM) as a pre-training task for the optical flow estimation (MatchFlow) with better feature representations, as GIM shares some common challenges as optical flow estimation, and with massive labeled real-world data. Thus, matching static scenes helps to learn more fundamental feature correlations of objects and scenes with consistent displacements. Specifically, the proposed MatchFlow model employs a QuadTree attention-based network pre-trained on MegaDepth to extract coarse features for further flow regression. Extensive experiments show that our model has great cross-dataset generalization. Our method achieves 11.5% and 10.1% error reduction from GMA on Sintel clean pass and KITTI test set. At the time of anonymous submission, our MatchFlow(G) enjoys state-of-theart performance on Sintel clean and final pass compared to published approaches with comparable computation and memory footprint. Codes and models will be released in https://github.com/DQiaole/MatchFlow.

Robust principal component analysis (RPCA) is widely studied in computer vision. Recently an adaptive rank estimate based RPCA has achieved top performance in low-level vision tasks without the prior rank, but both the rank estimate and RPCA optimization algorithm involve singular value decomposition, which requires extremely huge computational resource for large-scale matrices. To address these issues, an efficient RPCA (eRPCA) algorithm is proposed based on block Krylov iteration and CUR decomposition in this paper. Specifically, the Krylov iteration method is employed to approximate the eigenvalue decomposition in the rank estimation, which requires O(ndrq + n(rq)^2) for an (n×d) input matrix, in which q is a parameter with a small value, r is the target rank. Based on the estimated rank, CUR decomposition is adopted to replace SVD in updating low-rank matrix component, whose complexity reduces from O(rnd) to O(r^2n) per iteration. Experimental results verify the efficiency and effectiveness of the proposed eRPCA over the state-of-the-art methods in various low-level vision applications.

Vector graphics (VG) are ubiquitous in industrial designs. In this paper, we address semantic segmentation of a typical VG, i.e., roughcast floorplans with bare wall structures, whose output can be directly used for further applications like interior furnishing and room space modeling. Previous semantic segmentation works mostly process well-decorated floorplans in raster images and usually yield aliased boundaries and outlier fragments in segmented rooms, due to pixel-level segmentation that ignores the regular elements (e.g. line segments) in vector floorplans. To overcome these issues, we propose to fully utilize the regular elements in vector floorplans for more integral segmentation. Our pipeline predicts room segmentation from vector floorplans by dually classifying line segments as room boundaries, and regions partitioned by line segments as room segments. To fully exploit the structural relationships between lines and regions, we use two-stream graph neural networks to process the line segments and partitioned regions respectively, and devise a novel modulated graph attention layer to fuse the heterogeneous information from one stream to the other. Extensive experiments show that by directly operating on vector floorplans, we outperform image-based methods in both mIoU and mAcc. In addition, we propose a new metric that captures room integrity and boundary regularity, …

Vision-centric joint perception and prediction (PnP) has become an emerging trend in autonomous driving research. It predicts the future states of the traffic participants in the surrounding environment from raw RGB images. However, it is still a critical challenge to synchronize features obtained at multiple camera views and timestamps due to inevitable geometric distortions and further exploit those spatial-temporal features. To address this issue, we propose a temporal bird’s-eye-view pyramid transformer (TBP-Former) for vision-centric PnP, which includes two novel designs. First, a pose-synchronized BEV encoder is proposed to map raw image inputs with any camera pose at any time to a shared and synchronized BEV space for better spatial-temporal synchronization. Second, a spatial-temporal pyramid transformer is introduced to comprehensively extract multi-scale BEV features and predict future BEV states with the support of spatial priors. Extensive experiments on nuScenes dataset show that our proposed framework overall outperforms all state-of-the-art vision-based prediction methods.

A self-driving vehicle (SDV) must be able to perceive its surroundings and predict the future behavior of other traffic participants. Existing works either perform object detection followed by trajectory forecasting of the detected objects, or predict dense occupancy and flow grids for the whole scene. The former poses a safety concern as the number of detections needs to be kept low for efficiency reasons, sacrificing object recall. The latter is computationally expensive due to the high-dimensionality of the output grid, and suffers from the limited receptive field inherent to fully convolutional networks. Furthermore, both approaches employ many computational resources predicting areas or objects that might never be queried by the motion planner. This motivates our unified approach to perception and future prediction that implicitly represents occupancy and flow over time with a single neural network. Our method avoids unnecessary computation, as it can be directly queried by the motion planner at continuous spatio-temporal locations. Moreover, we design an architecture that overcomes the limited receptive field of previous explicit occupancy prediction methods by adding an efficient yet effective global attention mechanism. Through extensive experiments in both urban and highway settings, we demonstrate that our implicit model outperforms the current state-of-the-art. For …

Rigorously testing autonomy systems is essential for making safe self-driving vehicles (SDV) a reality. It requires one to generate safety critical scenarios beyond what can be collected safely in the world, as many scenarios happen rarely on our roads. To accurately evaluate performance, we need to test the SDV on these scenarios in closed-loop, where the SDV and other actors interact with each other at each timestep. Previously recorded driving logs provide a rich resource to build these new scenarios from, but for closed loop evaluation, we need to modify the sensor data based on the new scene configuration and the SDV’s decisions, as actors might be added or removed and the trajectories of existing actors and the SDV will differ from the original log. In this paper, we present UniSim, a neural sensor simulator that takes a single recorded log captured by a sensor-equipped vehicle and converts it into a realistic closed-loop multi-sensor simulation. UniSim builds neural feature grids to reconstruct both the static background and dynamic actors in the scene, and composites them together to simulate LiDAR and camera data at new viewpoints, with actors added or removed and at new placements. To better handle extrapolated views, we …

Predicting the future trajectories of the traffic agents is a gordian technique in autonomous driving. However, trajectory prediction suffers from data imbalance in the prevalent datasets, and the tailed data is often more complicated and safety-critical. In this paper, we focus on dealing with the long-tail phenomenon in trajectory prediction. Previous methods dealing with long-tail data did not take into account the variety of motion patterns in the tailed data. In this paper, we put forward a future enhanced contrastive learning framework to recognize tail trajectory patterns and form a feature space with separate pattern clusters.Furthermore, a distribution aware hyper predictor is brought up to better utilize the shaped feature space.Our method is a model-agnostic framework and can be plugged into many well-known baselines. Experimental results show that our framework outperforms the state-of-the-art long-tail prediction method on tailed samples by 9.5% on ADE and 8.5% on FDE, while maintaining or slightly improving the averaged performance. Our method also surpasses many long-tail techniques on trajectory prediction task.

Learning to predict agent motions with relationship reasoning is important for many applications. In motion prediction tasks, maintaining motion equivariance under Euclidean geometric transformations and invariance of agent interaction is a critical and fundamental principle. However, such equivariance and invariance properties are overlooked by most existing methods. To fill this gap, we propose EqMotion, an efficient equivariant motion prediction model with invariant interaction reasoning. To achieve motion equivariance, we propose an equivariant geometric feature learning module to learn a Euclidean transformable feature through dedicated designs of equivariant operations. To reason agent’s interactions, we propose an invariant interaction reasoning module to achieve a more stable interaction modeling. To further promote more comprehensive motion features, we propose an invariant pattern feature learning module to learn an invariant pattern feature, which cooperates with the equivariant geometric feature to enhance network expressiveness. We conduct experiments for the proposed model on four distinct scenarios: particle dynamics, molecule dynamics, human skeleton motion prediction and pedestrian trajectory prediction. Experimental results show that our method is not only generally applicable, but also achieves state-of-the-art prediction performances on all the four tasks, improving by 24.0/30.1/8.6/9.2%. Code is available at https://github.com/MediaBrain-SJTU/EqMotion.

We propose lookahead diffusion probabilistic models (LA-DPMs) to exploit the correlation in the outputs of the deep neural networks (DNNs) over subsequent timesteps in diffusion probabilistic models (DPMs) to refine the mean estimation of the conditional Gaussian distributions in the backward process. A typical DPM first obtains an estimate of the original data sample x by feeding the most recent state zi and index i into the DNN model and then computes the mean vector of the conditional Gaussian distribution for z{i-1}. We propose to calculate a more accurate estimate for x by performing extrapolation on the two estimates of x that are obtained by feeding (z{i+1}, i+1) and (zi, i) into the DNN model. The extrapolation can be easily integrated into the backward process of existing DPMs by introducing an additional connection over two consecutive timesteps, and fine-tuning is not required. Extensive experiments showed that plugging in the additional connection into DDPM, DDIM, DEIS, S-PNDM, and high-order DPM-Solvers leads to a significant performance gain in terms of Fréchet inception distance (FID) score. Our implementation is available at https://github.com/guoqiangzhang-x/LA-DPM.

Legged robots have the potential to expand the reach of autonomy beyond paved roads. In this work, we consider the difficult problem of locomotion on challenging terrains using a single forward-facing depth camera. Due to the partial observability of the problem, the robot has to rely on past observations to infer the terrain currently beneath it. To solve this problem, we follow the paradigm in computer vision that explicitly models the 3D geometry of the scene and propose Neural Volumetric Memory (NVM), a geometric memory architecture that explicitly accounts for the SE(3) equivariance of the 3D world. NVM aggregates feature volumes from multiple camera views by first bringing them back to the ego-centric frame of the robot. We test the learned visual-locomotion policy on a physical robot and show that our approach, learning legged locomotion with neural volumetric memory, produces performance gains over prior works on challenging terrains. We include ablation studies and show that the representations stored in the neural volumetric memory capture sufficient geometric information to reconstruct the scene. Our project page with videos is https://rchalyang.github.io/NVM/

Predicting human gaze is important in Human-Computer Interaction (HCI). However, to practically serve HCI applications, gaze prediction models must be scalable, fast, and accurate in their spatial and temporal gaze predictions. Recent scanpath prediction models focus on goal-directed attention (search). Such models are limited in their application due to a common approach relying on trained target detectors for all possible objects, and the availability of human gaze data for their training (both not scalable). In response, we pose a new task called ZeroGaze, a new variant of zero-shot learning where gaze is predicted for never-before-searched objects, and we develop a novel model, Gazeformer, to solve the ZeroGaze problem. In contrast to existing methods using object detector modules, Gazeformer encodes the target using a natural language model, thus leveraging semantic similarities in scanpath prediction. We use a transformer-based encoder-decoder architecture because transformers are particularly useful for generating contextual representations. Gazeformer surpasses other models by a large margin (19% - 70%) on the ZeroGaze setting. It also outperforms existing target-detection models on standard gaze prediction for both target-present and target-absent search tasks. In addition to its improved performance, Gazeformer is more than five times faster than the state-of-the-art target-present visual search model.

Recent approaches to drape garments quickly over arbitrary human bodies leverage self-supervision to eliminate the need for large training sets. However, they are designed to train one network per clothing item, which severely limits their generalization abilities. In our work, we rely on self-supervision to train a single network to drape multiple garments. This is achieved by predicting a 3D deformation field conditioned on the latent codes of a generative network, which models garments as unsigned distance fields. Our pipeline can generate and drape previously unseen garments of any topology, whose shape can be edited by manipulating their latent codes. Being fully differentiable, our formulation makes it possible to recover accurate 3D models of garments from partial observations -- images or 3D scans -- via gradient descent. Our code is publicly available at https://github.com/liren2515/DrapeNet.

Most existing multiple object tracking (MOT) methods that solely rely on appearance features struggle in tracking highly deformable objects. Other MOT methods that use motion clues to associate identities across frames have difficulty handling egocentric videos effectively or efficiently. In this work, we propose DETracker, a new MOT method that jointly detects and tracks deformable objects in egocentric videos. DETracker uses three novel modules, namely the motion disentanglement network (MDN), the patch association network (PAN) and the patch memory network (PMN), to explicitly tackle the difficulties caused by severe ego motion and fast morphing target objects. DETracker is end-to-end trainable and achieves near real-time speed. We also present DogThruGlasses, a large-scale deformable multi-object tracking dataset, with 150 videos and 73K annotated frames, collected by smart glasses. DETracker outperforms existing state-of-the-art method on the DogThruGlasses dataset and YouTube-Hand dataset.

Entangled representation of clothing and identity (ID)-intrinsic clues are potentially concomitant in conventional person Re-IDentification (ReID). Nevertheless, eliminating the negative impact of clothing on ID remains challenging due to the lack of theory and the difficulty of isolating the exact implications. In this paper, a causality-based Auto-Intervention Model, referred to as AIM, is first proposed to mitigate clothing bias for robust cloth-changing person ReID (CC-ReID). Specifically, we analyze the effect of clothing on the model inference and adopt a dual-branch model to simulate causal intervention. Progressively, clothing bias is eliminated automatically with model training. AIM is encouraged to learn more discriminative ID clues that are free from clothing bias. Extensive experiments on two standard CC-ReID datasets demonstrate the superiority of the proposed AIM over other state-of-the-art methods.

Micro-expression recognition is one of the most challenging topics in affective computing. It aims to recognize tiny facial movements difficult for humans to perceive in a brief period, i.e., 0.25 to 0.5 seconds. Recent advances in pre-training deep Bidirectional Transformers (BERT) have significantly improved self-supervised learning tasks in computer vision. However, the standard BERT in vision problems is designed to learn only from full images or videos, and the architecture cannot accurately detect details of facial micro-expressions. This paper presents Micron-BERT (µ-BERT), a novel approach to facial micro-expression recognition. The proposed method can automatically capture these movements in an unsupervised manner based on two key ideas. First, we employ Diagonal Micro-Attention (DMA) to detect tiny differences between two frames. Second, we introduce a new Patch of Interest (PoI) module to localize and highlight micro-expression interest regions and simultaneously reduce noisy backgrounds and distractions. By incorporating these components into an end-to-end deep network, the proposed µ-BERT significantly outperforms all previous work in various micro-expression tasks. µ-BERT can be trained on a large-scale unlabeled dataset, i.e., up to 8 million images, and achieves high accuracy on new unseen facial micro-expression datasets. Empirical experiments show µ-BERT consistently outperforms state-of-the-art performance on four micro-expression …

This paper proposes a self-supervised approach to learn universal facial representations from videos, that can transfer across a variety of facial analysis tasks such as Facial Attribute Recognition (FAR), Facial Expression Recognition (FER), DeepFake Detection (DFD), and Lip Synchronization (LS). Our proposed framework, named MARLIN, is a facial video masked autoencoder, that learns highly robust and generic facial embeddings from abundantly available non-annotated web crawled facial videos. As a challenging auxiliary task, MARLIN reconstructs the spatio-temporal details of the face from the densely masked facial regions which mainly include eyes, nose, mouth, lips, and skin to capture local and global aspects that in turn help in encoding generic and transferable features. Through a variety of experiments on diverse downstream tasks, we demonstrate MARLIN to be an excellent facial video encoder as well as feature extractor, that performs consistently well across a variety of downstream tasks including FAR (1.13% gain over supervised benchmark), FER (2.64% gain over unsupervised benchmark), DFD (1.86% gain over unsupervised benchmark), LS (29.36% gain for Frechet Inception Distance), and even in low data regime. Our code and models are available at https://github.com/ControlNet/MARLIN.

Despite recent advances in syncing lip movements with any audio waves, current methods still struggle to balance generation quality and the model’s generalization ability. Previous studies either require long-term data for training or produce a similar movement pattern on all subjects with low quality. In this paper, we propose StyleSync, an effective framework that enables high-fidelity lip synchronization. We identify that a style-based generator would sufficiently enable such a charming property on both one-shot and few-shot scenarios. Specifically, we design a mask-guided spatial information encoding module that preserves the details of the given face. The mouth shapes are accurately modified by audio through modulated convolutions. Moreover, our design also enables personalized lip-sync by introducing style space and generator refinement on only limited frames. Thus the identity and talking style of a target person could be accurately preserved. Extensive experiments demonstrate the effectiveness of our method in producing high-fidelity results on a variety of scenes.

Objects make unique sounds under different perturbations, environment conditions, and poses relative to the listener. While prior works have modeled impact sounds and sound propagation in simulation, we lack a standard dataset of impact sound fields of real objects for audio-visual learning and calibration of the sim-to-real gap. We present RealImpact, a large-scale dataset of real object impact sounds recorded under controlled conditions. RealImpact contains 150,000 recordings of impact sounds of 50 everyday objects with detailed annotations, including their impact locations, microphone locations, contact force profiles, material labels, and RGBD images. We make preliminary attempts to use our dataset as a reference to current simulation methods for estimating object impact sounds that match the real world. Moreover, we demonstrate the usefulness of our dataset as a testbed for acoustic and audio-visual learning via the evaluation of two benchmark tasks, including listener location classification and visual acoustic matching.

We study the problem of human action recognition using motion capture (MoCap) sequences. Unlike existing techniques that take multiple manual steps to derive standardized skeleton representations as model input, we propose a novel Spatial-Temporal Mesh Transformer (STMT) to directly model the mesh sequences. The model uses a hierarchical transformer with intra-frame off-set attention and inter-frame self-attention. The attention mechanism allows the model to freely attend between any two vertex patches to learn non-local relationships in the spatial-temporal domain. Masked vertex modeling and future frame prediction are used as two self-supervised tasks to fully activate the bi-directional and auto-regressive attention in our hierarchical transformer. The proposed method achieves state-of-the-art performance compared to skeleton-based and point-cloud-based models on common MoCap benchmarks. Code is available at https://github.com/zgzxy001/STMT.

In this paper, we propose an efficient event-based motion estimation framework for various motion models. Different from previous works, we design a progressive event-to-map alignment scheme and utilize the spatio-temporal correlations to align events. In detail, we progressively align sampled events in an event batch to the time-surface map and obtain the updated motion model by minimizing a novel time-surface loss. In addition, a dynamic batch size strategy is applied to adaptively adjust the batch size so that all events in the batch are consistent with the current motion model. Our framework has three advantages: a) the progressive scheme refines motion parameters iteratively, achieving accurate motion estimation; b) within one iteration, only a small portion of events are involved in optimization, which greatly reduces the total runtime; c) the dynamic batch size strategy ensures that the constant velocity assumption always holds. We conduct comprehensive experiments to evaluate our framework on challenging high-speed scenes with three motion models: rotational, homography, and 6-DOF models. Experimental results demonstrate that our framework achieves state-of-the-art estimation accuracy and efficiency.

State-of-the-art solutions for Shape-from-Polarization (SfP) suffer from a speed-resolution tradeoff: they either sacrifice the number of polarization angles measured or necessitate lengthy acquisition times due to framerate constraints, thus compromising either accuracy or latency. We tackle this tradeoff using event cameras. Event cameras operate at microseconds resolution with negligible motion blur, and output a continuous stream of events that precisely measures how light changes over time asynchronously. We propose a setup that consists of a linear polarizer rotating at high speeds in front of an event camera. Our method uses the continuous event stream caused by the rotation to reconstruct relative intensities at multiple polarizer angles. Experiments demonstrate that our method outperforms physics-based baselines using frames, reducing the MAE by 25% in synthetic and real-world datasets. In the real world, we observe, however, that the challenging conditions (i.e., when few events are generated) harm the performance of physics-based solutions. To overcome this, we propose a learning-based approach that learns to estimate surface normals even at low event-rates, improving the physics-based approach by 52% on the real world dataset. The proposed system achieves an acquisition speed equivalent to 50 fps (>twice the framerate of the commercial polarization sensor) while retaining the …

Event cameras sense the intensity changes asynchronously and produce event streams with high dynamic range and low latency. This has inspired research endeavors utilizing events to guide the challenging video super-resolution (VSR) task. In this paper, we make the first at tempt to address a novel problem of achieving VSR at random scales by taking advantages of the high temporal resolution property of events. This is hampered by the difficulties of representing the spatial-temporal information of events when guiding VSR. To this end, we propose a novel framework that incorporates the spatial-temporal interpolation of events to VSR in a unified framework. Our key idea is to learn implicit neural representations from queried spatial-temporal coordinates and features from both RGB frames and events. Our method contains three parts. Specifically, the Spatial-Temporal Fusion (STF) module first learns the 3D features from events and RGB frames. Then, the Temporal Filter (TF) module unlocks more explicit motion information from the events near the queried timestamp and generates the 2D features. Lastly, the Spatial-Temporal Implicit Representation (STIR) module recovers the SR frame in arbitrary resolutions from the outputs of these two modules. In addition, we collect a real-world dataset with spatially aligned events and RGB …

A novel 4K video frame interpolator based on bilateral transformer (BiFormer) is proposed in this paper, which performs three steps: global motion estimation, local motion refinement, and frame synthesis. First, in global motion estimation, we predict symmetric bilateral motion fields at a coarse scale. To this end, we propose BiFormer, the first transformer-based bilateral motion estimator. Second, we refine the global motion fields efficiently using blockwise bilateral cost volumes (BBCVs). Third, we warp the input frames using the refined motion fields and blend them to synthesize an intermediate frame. Extensive experiments demonstrate that the proposed BiFormer algorithm achieves excellent interpolation performance on 4K datasets. The source codes are available at https://github.com/JunHeum/BiFormer.

Flow-guided synthesis provides a common framework for frame interpolation, where optical flow is estimated to guide the synthesis of intermediate frames between consecutive inputs. In this paper, we present UPR-Net, a novel Unified Pyramid Recurrent Network for frame interpolation. Cast in a flexible pyramid framework, UPR-Net exploits lightweight recurrent modules for both bi-directional flow estimation and intermediate frame synthesis. At each pyramid level, it leverages estimated bi-directional flow to generate forward-warped representations for frame synthesis; across pyramid levels, it enables iterative refinement for both optical flow and intermediate frame. In particular, we show that our iterative synthesis strategy can significantly improve the robustness of frame interpolation on large motion cases. Despite being extremely lightweight (1.7M parameters), our base version of UPR-Net achieves excellent performance on a large range of benchmarks. Code and trained models of our UPR-Net series are available at: https://github.com/srcn-ivl/UPR-Net.

Restoring sharp high frame-rate videos from low frame-rate blurry videos is a challenging problem. Existing blurry frame interpolation methods assume a predefined and known exposure time, which suffer from severe performance drop when applied to videos captured in the wild. In this paper, we study the problem of blurry frame interpolation under blind exposure with the assistance of an event camera. The high temporal resolution of the event camera is beneficial to obtain the exposure prior that is lost during the imaging process. Besides, sharp frames can be restored using event streams and blurry frames relying on the mutual constraint among them. Therefore, we first propose an exposure estimation strategy guided by event streams to estimate the lost exposure prior, transforming the blind exposure problem well-posed. Second, we propose to model the mutual constraint with a temporal-exposure control strategy through iterative residual learning. Our blurry frame interpolation method achieves a distinct performance boost over existing methods on both synthetic and self-collected real-world datasets under blind exposure.

FlowFormer introduces a transformer architecture into optical flow estimation and achieves state-of-the-art performance. The core component of FlowFormer is the transformer-based cost-volume encoder. Inspired by recent success of masked autoencoding (MAE) pretraining in unleashing transformers’ capacity of encoding visual representation, we propose Masked Cost Volume Autoencoding (MCVA) to enhance FlowFormer by pretraining the cost-volume encoder with a novel MAE scheme. Firstly, we introduce a block-sharing masking strategy to prevent masked information leakage, as the cost maps of neighboring source pixels are highly correlated. Secondly, we propose a novel pre-text reconstruction task, which encourages the cost-volume encoder to aggregate long-range information and ensures pretraining-finetuning consistency. We also show how to modify the FlowFormer architecture to accommodate masks during pretraining. Pretrained with MCVA, our proposed FlowFormer++ ranks 1st among published methods on both Sintel and KITTI-2015 benchmarks. Specifically, FlowFormer++ achieves 1.07 and 1.94 average end-point-error (AEPE) on the clean and final pass of Sintel benchmark, leading to 7.76% and 7.18% error reductions from FlowFormer. FlowFormer++ obtains 4.52 F1-all on the KITTI-2015 test set, improving FlowFormer by 0.16.

Transformer architectures have achieved SOTA performance on the human mesh recovery (HMR) from monocular images. However, the performance gain has come at the cost of substantial memory and computational overhead. A lightweight and efficient model to reconstruct accurate human mesh is needed for real-world applications. In this paper, we propose a pure transformer architecture named POoling aTtention TransformER (POTTER) for the HMR task from single images. Observing that the conventional attention module is memory and computationally expensive, we propose an efficient pooling attention module, which significantly reduces the memory and computational cost without sacrificing performance. Furthermore, we design a new transformer architecture by integrating a High-Resolution (HR) stream for the HMR task. The high-resolution local and global features from the HR stream can be utilized for recovering more accurate human mesh. Our POTTER outperforms the SOTA method METRO by only requiring 7% of total parameters and 14% of the Multiply-Accumulate Operations on the Human3.6M (PA-MPJPE) and 3DPW (all three metrics) datasets. Code will be publicly available.

In recent years, deep learning-based approaches have shown great strength in multi-view stereo because of their outstanding ability to extract robust visual features. However, most learning-based methods need to build the cost volume and increase the receptive field enormously to get a satisfactory result when dealing with large-scale textureless regions, consequently leading to prohibitive memory consumption. To ensure both memory-friendly and textureless-resilient, we innovatively transplant the spirit of deformable convolution from deep learning into the traditional PatchMatch-based method. Specifically, for each pixel with matching ambiguity (termed unreliable pixel), we adaptively deform the patch centered on it to extend the receptive field until covering enough correlative reliable pixels (without matching ambiguity) that serve as anchors. When performing PatchMatch, constrained by the anchor pixels, the matching cost of an unreliable pixel is guaranteed to reach the global minimum at the correct depth and therefore increases the robustness of multi-view stereo significantly. To detect more anchor pixels to ensure better adaptive patch deformation, we propose to evaluate the matching ambiguity of a certain pixel by checking the convergence of the estimated depth as optimization proceeds. As a result, our method achieves state-of-the-art performance on ETH3D and Tanks and Temples while preserving low memory …

Explicit visible videos can provide sufficient visual information and facilitate vision applications. Unfortunately, the image sensors of visible cameras are sensitive to light conditions like darkness or overexposure. To make up for this, recently, infrared sensors capable of stable imaging have received increasing attention in autonomous driving and monitoring. However, most prosperous vision models are still trained on massive clear visible data, facing huge visual gaps when deploying to infrared imaging scenarios. In such cases, transferring the infrared video to a distinct visible one with fine-grained semantic patterns is a worthwhile endeavor. Previous works improve the outputs by equally optimizing each patch on the translated visible results, which is unfair for enhancing the details on content-rich patches due to the long-tail effect of pixel distribution. Here we propose a novel CPTrans framework to tackle the challenge via balancing gradients of different patches, achieving the fine-grained Content-rich Patches Transferring. Specifically, the content-aware optimization module encourages model optimization along gradients of target patches, ensuring the improvement of visual details. Additionally, the content-aware temporal normalization module enforces the generator to be robust to the motions of target patches. Moreover, we extend the existing dataset InfraredCity to more challenging adverse weather conditions (rain and …

Robust and non-destructive material classification is a challenging but crucial first-step in numerous vision applications. We propose a physics-guided material classification framework that relies on thermal properties of the object. Our key observation is that the rate of heating and cooling of an object depends on the unique intrinsic properties of the material, namely the emissivity and diffusivity. We leverage this observation by gently heating the objects in the scene with a low-power laser for a fixed duration and then turning it off, while a thermal camera captures measurements during the heating and cooling process. We then take this spatial and temporal “thermal spread function” (TSF) to solve an inverse heat equation using the finite-differences approach, resulting in a spatially varying estimate of diffusivity and emissivity. These tuples are then used to train a classifier that produces a fine-grained material label at each spatial pixel. Our approach is extremely simple requiring only a small light source (low power laser) and a thermal camera, and produces robust classification results with 86% accuracy over 16 classes

Most of the existing blind image Super-Resolution (SR) methods assume that the blur kernels are space-invariant. However, the blur involved in real applications are usually space-variant due to object motion, out-of-focus, etc., resulting in severe performance drop of the advanced SR methods. To address this problem, we firstly introduce two new datasets with out-of-focus blur, i.e., NYUv2-BSR and Cityscapes-BSR, to support further researches of blind SR with space-variant blur. Based on the datasets, we design a novel Cross-MOdal fuSion network (CMOS) that estimate both blur and semantics simultaneously, which leads to improved SR results. It involves a feature Grouping Interactive Attention (GIA) module to make the two modals interact more effectively and avoid inconsistency. GIA can also be used for the interaction of other features because of the universality of its structure. Qualitative and quantitative experiments compared with state-of-the-art methods on above datasets and real-world images demonstrate the superiority of our method, e.g., obtaining PSNR/SSIM by +1.91/+0.0048 on NYUv2-BSR than MANet.

Low-light image enhancement (LLIE) investigates how to improve illumination and produce normal-light images. The majority of existing methods improve low-light images via a global and uniform manner, without taking into account the semantic information of different regions. Without semantic priors, a network may easily deviate from a region’s original color. To address this issue, we propose a novel semantic-aware knowledge-guided framework (SKF) that can assist a low-light enhancement model in learning rich and diverse priors encapsulated in a semantic segmentation model. We concentrate on incorporating semantic knowledge from three key aspects: a semantic-aware embedding module that wisely integrates semantic priors in feature representation space, a semantic-guided color histogram loss that preserves color consistency of various instances, and a semantic-guided adversarial loss that produces more natural textures by semantic priors. Our SKF is appealing in acting as a general framework in LLIE task. Extensive experiments show that models equipped with the SKF significantly outperform the baselines on multiple datasets and our SKF generalizes to different models and scenes well. The code is available at Semantic-Aware-Low-Light-Image-Enhancement.

Data augmentation (DA) is an efficient strategy for improving the performance of deep neural networks. Recent DA strategies have demonstrated utility in single image super-resolution (SR). Little research has, however, focused on the DA strategy for light field SR, in which multi-view information utilization is required. For the first time in light field SR, we propose a potent DA strategy called CutMIB to improve the performance of existing light field SR networks while keeping their structures unchanged. Specifically, CutMIB first cuts low-resolution (LR) patches from each view at the same location. Then CutMIB blends all LR patches to generate the blended patch and finally pastes the blended patch to the corresponding regions of high-resolution light field views, and vice versa. By doing so, CutMIB enables light field SR networks to learn from implicit geometric information during the training stage. Experimental results demonstrate that CutMIB can improve the reconstruction performance and the angular consistency of existing light field SR networks. We further verify the effectiveness of CutMIB on real-world light field SR and light field denoising. The implementation code is available at https://github.com/zeyuxiao1997/CutMIB.

Modeling and synthesizing real noise in the standard RGB (sRGB) domain is challenging due to the complicated noise distribution. While most of the deep noise generators proposed to synthesize sRGB real noise using an end-to-end trained model, the lack of explicit noise modeling degrades the quality of their synthesized noise. In this work, we propose to model the real noise as not only dependent on the underlying clean image pixel intensity, but also highly correlated to its neighboring noise realization within the local region. Correspondingly, we propose a novel noise synthesizing framework by explicitly learning its neighboring correlation on top of the signal dependency. With the proposed noise model, our framework greatly bridges the distribution gap between synthetic noise and real noise. We show that our generated “real” sRGB noisy images can be used for training supervised deep denoisers, thus to improve their real denoising results with a large margin, comparing to the popular classic denoisers or the deep denoisers that are trained on other sRGB noise generators. The code will be available at https://github.com/xuan611/sRGB-Real-Noise-Synthesizing.

When capturing and storing images, devices inevitably introduce noise. Reducing this noise is a critical task called image denoising. Deep learning has become the de facto method for image denoising, especially with the emergence of Transformer-based models that have achieved notable state-of-the-art results on various image tasks. However, deep learning-based methods often suffer from a lack of generalization ability. For example, deep models trained on Gaussian noise may perform poorly when tested on other noise distributions. To address this issue, we present a novel approach to enhance the generalization performance of denoising networks, known as masked training. Our method involves masking random pixels of the input image and reconstructing the missing information during training. We also mask out the features in the self-attention layers to avoid the impact of training-testing inconsistency. Our approach exhibits better generalization ability than other deep learning models and is directly applicable to real-world scenarios. Additionally, our interpretability analysis demonstrates the superiority of our method.

Blind face restoration usually synthesizes degraded low-quality data with a pre-defined degradation model for training, while more complex cases could happen in the real world. This gap between the assumed and actual degradation hurts the restoration performance where artifacts are often observed in the output. However, it is expensive and infeasible to include every type of degradation to cover real-world cases in the training data. To tackle this robustness issue, we propose Diffusion-based Robust Degradation Remover (DR2) to first transform the degraded image to a coarse but degradation-invariant prediction, then employ an enhancement module to restore the coarse prediction to a high-quality image. By leveraging a well-performing denoising diffusion probabilistic model, our DR2 diffuses input images to a noisy status where various types of degradation give way to Gaussian noise, and then captures semantic information through iterative denoising steps. As a result, DR2 is robust against common degradation (e.g. blur, resize, noise and compression) and compatible with different designs of enhancement modules. Experiments in various settings show that our framework outperforms state-of-the-art methods on heavily degraded synthetic and real-world datasets.

In recent years, we have witnessed the great advancement of Deep neural networks (DNNs) in image restoration. However, a critical limitation is that they cannot generalize well to real-world degradations with different degrees or types. In this paper, we are the first to propose a novel training strategy for image restoration from the causality perspective, to improve the generalization ability of DNNs for unknown degradations. Our method, termed Distortion Invariant representation Learning (DIL), treats each distortion type and degree as one specific confounder, and learns the distortion-invariant representation by eliminating the harmful confounding effect of each degradation. We derive our DIL with the back-door criterion in causality by modeling the interventions of different distortions from the optimization perspective. Particularly, we introduce counterfactual distortion augmentation to simulate the virtual distortion types and degrees as the confounders. Then, we instantiate the intervention of each distortion with a virtual model updating based on corresponding distorted images, and eliminate them from the meta-learning perspective. Extensive experiments demonstrate the generalization capability of our DIL on unseen distortion types and degrees. Our code will be available at https://github.com/lixinustc/Causal-IR-DIL.

Single-image super-resolution (SISR) networks trained with perceptual and adversarial losses provide high-contrast outputs compared to those of networks trained with distortion-oriented losses, such as L1 or L2. However, it has been shown that using a single perceptual loss is insufficient for accurately restoring locally varying diverse shapes in images, often generating undesirable artifacts or unnatural details. For this reason, combinations of various losses, such as perceptual, adversarial, and distortion losses, have been attempted, yet it remains challenging to find optimal combinations. Hence, in this paper, we propose a new SISR framework that applies optimal objectives for each region to generate plausible results in overall areas of high-resolution outputs. Specifically, the framework comprises two models: a predictive model that infers an optimal objective map for a given low-resolution (LR) input and a generative model that applies a target objective map to produce the corresponding SR output. The generative model is trained over our proposed objective trajectory representing a set of essential objectives, which enables the single network to learn various SR results corresponding to combined losses on the trajectory. The predictive model is trained using pairs of LR images and corresponding optimal objective maps searched from the objective trajectory. Experimental results …

The popular VQ-VAE models reconstruct images through learning a discrete codebook but suffer from a significant issue in the rapid quality degradation of image reconstruction as the compression rate rises. One major reason is that a higher compression rate induces more loss of visual signals on the higher frequency spectrum, which reflect the details on pixel space. In this paper, a Frequency Complement Module (FCM) architecture is proposed to capture the missing frequency information for enhancing reconstruction quality. The FCM can be easily incorporated into the VQ-VAE structure, and we refer to the new model as Frequancy Augmented VAE (FA-VAE). In addition, a Dynamic Spectrum Loss (DSL) is introduced to guide the FCMs to balance between various frequencies dynamically for optimal reconstruction. FA-VAE is further extended to the text-to-image synthesis task, and a Cross-attention Autoregressive Transformer (CAT) is proposed to obtain more precise semantic attributes in texts. Extensive reconstruction experiments with different compression rates are conducted on several benchmark datasets, and the results demonstrate that the proposed FA-VAE is able to restore more faithfully the details compared to SOTA methods. CAT also shows improved generation quality with better image-text semantic alignment.

User-generated content (UGC) live videos are often bothered by various distortions during capture procedures and thus exhibit diverse visual qualities. Such source videos are further compressed and transcoded by media server providers before being distributed to end-users. Because of the flourishing of UGC live videos, effective video quality assessment (VQA) tools are needed to monitor and perceptually optimize live streaming videos in the distributing process. Unfortunately, existing compressed UGC VQA databases are either small in scale or employ high-quality UGC videos as source videos, so VQA models developed on these databases have limited abilities to evaluate UGC live videos. In this paper, we address UGC Live VQA problems by constructing a first-of-a-kind subjective UGC Live VQA database and developing an effective evaluation tool. Concretely, 418 source UGC videos are collected in real live streaming scenarios and 3,762 compressed ones at different bit rates are generated for the subsequent subjective VQA experiments. Based on the built database, we develop a Multi-Dimensional VQA (MD-VQA) evaluator to measure the visual quality of UGC live videos from semantic, distortion, and motion aspects respectively. Extensive experimental results show that MD-VQA achieves state-of-the-art performance on both our UGC Live VQA database and existing compressed UGC VQA …

With the development of high-definition display devices, the practical scenario of Super-Resolution (SR) usually needs to super-resolve large input like 2K to higher resolution (4K/8K). To reduce the computational and memory cost, current methods first split the large input into local patches and then merge the SR patches into the output. These methods adaptively allocate a subnet for each patch. Quantization is a very important technique for network acceleration and has been used to design the subnets. Current methods train an MLP bit selector to determine the propoer bit for each layer. However, they uniformly sample subnets for training, making simple subnets overfitted and complicated subnets underfitted. Therefore, the trained bit selector fails to determine the optimal bit. Apart from this, the introduced bit selector brings additional cost to each layer of the SR network. In this paper, we propose a novel method named Content-Aware Bit Mapping (CABM), which can remove the bit selector without any performance loss. CABM also learns a bit selector for each layer during training. After training, we analyze the relation between the edge information of an input patch and the bit of each layer. We observe that the edge information can be an effective metric …

In this paper, we examine gradients of logits of image classification CNNs by input pixel values. We observe that these fluctuate considerably with training randomness, such as the random initialization of the networks. We extend our study to gradients of intermediate layers, obtained via GradCAM, as well as popular network saliency estimators such as DeepLIFT, SHAP, LIME, Integrated Gradients, and SmoothGrad. While empirical noise levels vary, qualitatively different attributions to image features are still possible with all of these, which comes with implications for interpreting such attributions, in particular when seeking data-driven explanations of the phenomenon generating the data. Finally, we demonstrate that the observed artefacts can be removed by marginalization over the initialization distribution by simple stochastic integration.

Flow-based methods have demonstrated promising results in addressing the ill-posed nature of super-resolution (SR) by learning the distribution of high-resolution (HR) images with the normalizing flow. However, these methods can only perform a predefined fixed-scale SR, limiting their potential in real-world applications. Meanwhile, arbitrary-scale SR has gained more attention and achieved great progress. Nonetheless, previous arbitrary-scale SR methods ignore the ill-posed problem and train the model with per-pixel L1 loss, leading to blurry SR outputs. In this work, we propose “Local Implicit Normalizing Flow” (LINF) as a unified solution to the above problems. LINF models the distribution of texture details under different scaling factors with normalizing flow. Thus, LINF can generate photo-realistic HR images with rich texture details in arbitrary scale factors. We evaluate LINF with extensive experiments and show that LINF achieves the state-of-the-art perceptual quality compared with prior arbitrary-scale SR methods.

The ability of scale-equivariance processing blocks plays a central role in arbitrary-scale image super-resolution tasks. Inspired by this crucial observation, this work proposes two novel scale-equivariant modules within a transformer-style framework to enhance arbitrary-scale image super-resolution (ASISR) performance, especially in high upsampling rate image extrapolation. In the feature extraction phase, we design a plug-in module called Adaptive Feature Extractor, which injects explicit scale information in frequency-expanded encoding, thus achieving scale-adaption in representation learning. In the upsampling phase, a learnable Neural Kriging upsampling operator is introduced, which simultaneously encodes both relative distance (i.e., scale-aware) information as well as feature similarity (i.e., with priori learned from training data) in a bilateral manner, providing scale-encoded spatial feature fusion. The above operators are easily plugged into multiple stages of a SR network, and a recent emerging pre-training strategy is also adopted to impulse the model’s performance further. Extensive experimental results have demonstrated the outstanding scale-equivariance capability offered by the proposed operators and our learning framework, with much better results than previous SOTAs at arbitrary scales for SR. Our code is available at https://github.com/neuralchen/EQSR.

Learning continuous image representations is recently gaining popularity for image super-resolution (SR) because of its ability to reconstruct high-resolution images with arbitrary scales from low-resolution inputs. Existing methods mostly ensemble nearby features to predict the new pixel at any queried coordinate in the SR image. Such a local ensemble suffers from some limitations: i) it has no learnable parameters and it neglects the similarity of the visual features; ii) it has a limited receptive field and cannot ensemble relevant features in a large field which are important in an image. To address these issues, this paper proposes a continuous implicit attention-in-attention network, called CiaoSR. We explicitly design an implicit attention network to learn the ensemble weights for the nearby local features. Furthermore, we embed a scale-aware attention in this implicit attention network to exploit additional non-local information. Extensive experiments on benchmark datasets demonstrate CiaoSR significantly outperforms the existing single image SR methods with the same backbone. In addition, CiaoSR also achieves the state-of-the-art performance on the arbitrary-scale SR task. The effectiveness of the method is also demonstrated on the real-world SR setting. More importantly, CiaoSR can be flexibly integrated into any backbone to improve the SR performance.

We develop a simple yet surprisingly effective implicit representing scheme called Multiplicative Fourier Level of Detail (MFLOD) motivated by the recent success of multiplicative filter network. Built on multi-resolution feature grid/volume (e.g., the sparse voxel octree), each level’s feature is first modulated by a sinusoidal function and then element-wisely multiplied by a linear transformation of previous layer’s representation in a layer-to-layer recursive manner, yielding the scale-aggregated encodings for a subsequent simple linear forward to get final output. In contrast to previous hybrid representations relying on interleaved multilevel fusion and nonlinear activation-based decoding, MFLOD could be elegantly characterized as a linear combination of sine basis functions with varying amplitude, frequency, and phase upon the learned multilevel features, thus offering great feasibility in Fourier analysis. Comprehensive experimental results on implicit neural representation learning tasks including image fitting, 3D shape representation, and neural radiance fields well demonstrate the superior quality and generalizability achieved by the proposed MFLOD scheme.

Existing works on document image shadow removal mostly depend on learning and leveraging a constant background (the color of the paper) from the image. However, the constant background is less representative and frequently ignores other background colors, such as the printed colors, resulting in distorted results. In this paper, we present a color-aware background extraction network (CBENet) for extracting a spatially varying background image that accurately depicts the background colors of the document. Furthermore, we propose a background-guided document images shadow removal network (BGShadowNet) using the predicted spatially varying background as auxiliary information, which consists of two stages. At Stage I, a background-constrained decoder is designed to promote a coarse result. Then, the coarse result is refined with a background-based attention module (BAModule) to maintain a consistent appearance and a detail improvement module (DEModule) to enhance the texture details at Stage II. Experiments on two benchmark datasets qualitatively and quantitatively validate the superiority of the proposed approach over state-of-the-arts.

We present a novel image inversion framework and a training pipeline to achieve high-fidelity image inversion with high-quality attribute editing. Inverting real images into StyleGAN’s latent space is an extensively studied problem, yet the trade-off between the image reconstruction fidelity and image editing quality remains an open challenge. The low-rate latent spaces are limited in their expressiveness power for high-fidelity reconstruction. On the other hand, high-rate latent spaces result in degradation in editing quality. In this work, to achieve high-fidelity inversion, we learn residual features in higher latent codes that lower latent codes were not able to encode. This enables preserving image details in reconstruction. To achieve high-quality editing, we learn how to transform the residual features for adapting to manipulations in latent codes. We train the framework to extract residual features and transform them via a novel architecture pipeline and cycle consistency losses. We run extensive experiments and compare our method with state-of-the-art inversion methods. Qualitative metrics and visual comparisons show significant improvements.

We investigate the problem of automatically placing an object into a background image for image compositing. Given a background image and a segmented object, the goal is to train a model to predict plausible placements (location and scale) of the object for compositing. The quality of the composite image highly depends on the predicted location/scale. Existing works either generate candidate bounding boxes or apply sliding-window search using global representations from background and object images, which fail to model local information in background images. However, local clues in background images are important to determine the compatibility of placing the objects with certain locations/scales. In this paper, we propose to learn the correlation between object features and all local background features with a transformer module so that detailed information can be provided on all possible location/scale configurations. A sparse contrastive loss is further proposed to train our model with sparse supervision. Our new formulation generates a 3D heatmap indicating the plausibility of all location/scale combinations in one network forward pass, which is >10× faster than the previous sliding-window method. It also supports interactive search when users provide a pre-defined location or scale. The proposed method can be trained with explicit annotation or …

Font design is of vital importance in the digital content design and modern printing industry. Developing algorithms capable of automatically synthesizing vector fonts can significantly facilitate the font design process. However, existing methods mainly concentrate on raster image generation, and only a few approaches can directly synthesize vector fonts. This paper proposes an end-to-end trainable method, VecFontSDF, to reconstruct and synthesize high-quality vector fonts using signed distance functions (SDFs). Specifically, based on the proposed SDF-based implicit shape representation, VecFontSDF learns to model each glyph as shape primitives enclosed by several parabolic curves, which can be precisely converted to quadratic Bézier curves that are widely used in vector font products. In this manner, most image generation methods can be easily extended to synthesize vector fonts. Qualitative and quantitative experiments conducted on a publicly-available dataset demonstrate that our method obtains high-quality results on several tasks, including vector font reconstruction, interpolation, and few-shot vector font synthesis, markedly outperforming the state of the art.

Content and style disentanglement is an effective way to achieve few-shot font generation. It allows to transfer the style of the font image in a source domain to the style defined with a few reference images in a target domain. However, the content feature extracted using a representative font might not be optimal. In light of this, we propose a content fusion module (CFM) to project the content feature into a linear space defined by the content features of basis fonts, which can take the variation of content features caused by different fonts into consideration. Our method also allows to optimize the style representation vector of reference images through a lightweight iterative style-vector refinement (ISR) strategy. Moreover, we treat the 1D projection of a character image as a probability distribution and leverage the distance between two distributions as the reconstruction loss (namely projected character loss, PCL). Compared to L2 or L1 reconstruction loss, the distribution distance pays more attention to the global shape of characters. We have evaluated our method on a dataset of 300 fonts with 6.5k characters each. Experimental results verify that our method outperforms existing state-of-the-art few-shot font generation methods by a large margin. The source code …

Semantic image editing provides users with a flexible tool to modify a given image guided by a corresponding segmentation map. In this task, the features of the foreground objects and the backgrounds are quite different. However, all previous methods handle backgrounds and objects as a whole using a monolithic model. Consequently, they remain limited in processing content-rich images and suffer from generating unrealistic objects and texture-inconsistent backgrounds. To address this issue, we propose a novel paradigm, Semantic Image Editing by Disentangling Object and Background (SIEDOB), the core idea of which is to explicitly leverages several heterogeneous subnetworks for objects and backgrounds. First, SIEDOB disassembles the edited input into background regions and instance-level objects. Then, we feed them into the dedicated generators. Finally, all synthesized parts are embedded in their original locations and utilize a fusion network to obtain a harmonized result. Moreover, to produce high-quality edited images, we propose some innovative designs, including Semantic-Aware Self-Propagation Module, Boundary-Anchored Patch Discriminator, and Style-Diversity Object Generator, and integrate them into SIEDOB. We conduct extensive experiments on Cityscapes and ADE20K-Room datasets and exhibit that our method remarkably outperforms the baselines, especially in synthesizing realistic and diverse objects and texture-consistent backgrounds.

Recent conditional image generation methods produce images of remarkable diversity, fidelity and realism. However, the majority of these methods allow conditioning only on labels or text prompts, which limits their level of control over the generation result. In this paper, we introduce MaskSketch, an image generation method that allows spatial conditioning of the generation result using a guiding sketch as an extra conditioning signal during sampling. MaskSketch utilizes a pre-trained masked generative transformer, requiring no model training or paired supervision, and works with input sketches of different levels of abstraction. We show that intermediate self-attention maps of a masked generative transformer encode important structural information of the input image, such as scene layout and object shape, and we propose a novel sampling method based on this observation to enable structure-guided generation. Our results show that MaskSketch achieves high image realism and fidelity to the guiding structure. Evaluated on standard benchmark datasets, MaskSketch outperforms state-of-the-art methods for sketch-to-image translation, as well as unpaired image-to-image translation approaches. The code can be found on our project website: https://masksketch.github.io/

We propose Text2Scene, a method to automatically create realistic textures for virtual scenes composed of multiple objects. Guided by a reference image and text descriptions, our pipeline adds detailed texture on labeled 3D geometries in the room such that the generated colors respect the hierarchical structure or semantic parts that are often composed of similar materials. Instead of applying flat stylization on the entire scene at a single step, we obtain weak semantic cues from geometric segmentation, which are further clarified by assigning initial colors to segmented parts. Then we add texture details for individual objects such that their projections on image space exhibit feature embedding aligned with the embedding of the input. The decomposition makes the entire pipeline tractable to a moderate amount of computation resources and memory. As our framework utilizes the existing resources of image and text embedding, it does not require dedicated datasets with high-quality textures designed by skillful artists. To the best of our knowledge, it is the first practical and scalable approach that can create detailed and realistic textures of the desired style that maintain structural context for scenes with multiple objects.

Generative models have been widely studied in computer vision. Recently, diffusion models have drawn substantial attention due to the high quality of their generated images. A key desired property of image generative models is the ability to disentangle different attributes, which should enable modification towards a style without changing the semantic content, and the modification parameters should generalize to different images. Previous studies have found that generative adversarial networks (GANs) are inherently endowed with such disentanglement capability, so they can perform disentangled image editing without re-training or fine-tuning the network. In this work, we explore whether diffusion models are also inherently equipped with such a capability. Our finding is that for stable diffusion models, by partially changing the input text embedding from a neutral description (e.g., “a photo of person”) to one with style (e.g., “a photo of person with smile”) while fixing all the Gaussian random noises introduced during the denoising process, the generated images can be modified towards the target style without changing the semantic content. Based on this finding, we further propose a simple, light-weight image editing algorithm where the mixing weights of the two text embeddings are optimized for style matching and content preservation. This entire …

Diffusion models have shown impressive results in text-to-image synthesis. Using massive datasets of captioned images, diffusion models learn to generate raster images of highly diverse objects and scenes. However, designers frequently use vector representations of images like Scalable Vector Graphics (SVGs) for digital icons, graphics and stickers. Vector graphics can be scaled to any size, and are compact. In this work, we show that a text-conditioned diffusion model trained on pixel representations of images can be used to generate SVG-exportable vector graphics. We do so without access to large datasets of captioned SVGs. Instead, inspired by recent work on text-to-3D synthesis, we vectorize a text-to-image diffusion sample and fine-tune with a Score Distillation Sampling loss. By optimizing a differentiable vector graphics rasterizer, our method distills abstract semantic knowledge out of a pretrained diffusion model. By constraining the vector representation, we can also generate coherent pixel art and sketches. Our approach, VectorFusion, produces more coherent graphics than prior works that optimize CLIP, a contrastive image-text model.

Large-scale text-to-image generative models have been a revolutionary breakthrough in the evolution of generative AI, synthesizing diverse images with highly complex visual concepts. However, a pivotal challenge in leveraging such models for real-world content creation is providing users with control over the generated content. In this paper, we present a new framework that takes text-to-image synthesis to the realm of image-to-image translation -- given a guidance image and a target text prompt as input, our method harnesses the power of a pre-trained text-to-image diffusion model to generate a new image that complies with the target text, while preserving the semantic layout of the guidance image. Specifically, we observe and empirically demonstrate that fine-grained control over the generated structure can be achieved by manipulating spatial features and their self-attention inside the model. This results in a simple and effective approach, where features extracted from the guidance image are directly injected into the generation process of the translated image, requiring no training or fine-tuning. We demonstrate high-quality results on versatile text-guided image translation tasks, including translating sketches, rough drawings and animations into realistic images, changing the class and appearance of objects in a given image, and modifying global qualities such as lighting …

While generative models produce high-quality images of concepts learned from a large-scale database, a user often wishes to synthesize instantiations of their own concepts (for example, their family, pets, or items). Can we teach a model to quickly acquire a new concept, given a few examples? Furthermore, can we compose multiple new concepts together? We propose Custom Diffusion, an efficient method for augmenting existing text-to-image models. We find that only optimizing a few parameters in the text-to-image conditioning mechanism is sufficiently powerful to represent new concepts while enabling fast tuning (~6 minutes). Additionally, we can jointly train for multiple concepts or combine multiple fine-tuned models into one via closed-form constrained optimization. Our fine-tuned model generates variations of multiple new concepts and seamlessly composes them with existing concepts in novel settings. Our method outperforms or performs on par with several baselines and concurrent works in both qualitative and quantitative evaluations, while being memory and computationally efficient.

Layout generation aims to synthesize realistic graphic scenes consisting of elements with different attributes including category, size, position, and between-element relation. It is a crucial task for reducing the burden on heavy-duty graphic design works for formatted scenes, e.g., publications, documents, and user interfaces (UIs). Diverse application scenarios impose a big challenge in unifying various layout generation subtasks, including conditional and unconditional generation. In this paper, we propose a Layout Diffusion Generative Model (LDGM) to achieve such unification with a single decoupled diffusion model. LDGM views a layout of arbitrary missing or coarse element attributes as an intermediate diffusion status from a completed layout. Since different attributes have their individual semantics and characteristics, we propose to decouple the diffusion processes for them to improve the diversity of training samples and learn the reverse process jointly to exploit global-scope contexts for facilitating generation. As a result, our LDGM can generate layouts either from scratch or conditional on arbitrary available attributes. Extensive qualitative and quantitative experiments demonstrate our proposed LDGM outperforms existing layout generation models in both functionality and performance.

Image-to-image translation is an important and challenging problem in computer vision and image processing. Diffusion models(DM) have shown great potentials for high-quality image synthesis, and have gained competitive performance on the task of image-to-image translation. However, most of the existing diffusion models treat image-to-image translation as conditional generation processes, and suffer heavily from the gap between distinct domains. In this paper, a novel image-to-image translation method based on the Brownian Bridge Diffusion Model(BBDM) is proposed, which models image-to-image translation as a stochastic Brownian Bridge process, and learns the translation between two domains directly through the bidirectional diffusion process rather than a conditional generation process. To the best of our knowledge, it is the first work that proposes Brownian Bridge diffusion process for image-to-image translation. Experimental results on various benchmarks demonstrate that the proposed BBDM model achieves competitive performance through both visual inspection and measurable metrics.

Diffusion-based generative models have achieved remarkable success in image generation. Their guidance formulation allows an external model to plug-and-play control the generation process for various tasks without fine-tuning the diffusion model. However, the direct use of publicly available off-the-shelf models for guidance fails due to their poor performance on noisy inputs. For that, the existing practice is to fine-tune the guidance models with labeled data corrupted with noises. In this paper, we argue that this practice has limitations in two aspects: (1) performing on inputs with extremely various noises is too hard for a single guidance model; (2) collecting labeled datasets hinders scaling up for various tasks. To tackle the limitations, we propose a novel strategy that leverages multiple experts where each expert is specialized in a particular noise range and guides the reverse process of the diffusion at its corresponding timesteps. However, as it is infeasible to manage multiple networks and utilize labeled data, we present a practical guidance framework termed Practical Plug-And-Play (PPAP), which leverages parameter-efficient fine-tuning and data-free knowledge transfer. We exhaustively conduct ImageNet class conditional generation experiments to show that our method can successfully guide diffusion with small trainable parameters and no labeled data. Finally, we …

Denoising diffusion (score-based) generative models have recently achieved significant accomplishments in generating realistic and diverse data. These approaches define a forward diffusion process for transforming data into noise and a backward denoising process for sampling data from noise. Unfortunately, the generation process of current denoising diffusion models is notoriously slow due to the lengthy iterative noise estimations, which rely on cumbersome neural networks. It prevents the diffusion models from being widely deployed, especially on edge devices. Previous works accelerate the generation process of diffusion model (DM) via finding shorter yet effective sampling trajectories. However, they overlook the cost of noise estimation with a heavy network in every iteration. In this work, we accelerate generation from the perspective of compressing the noise estimation network. Due to the difficulty of retraining DMs, we exclude mainstream training-aware compression paradigms and introduce post-training quantization (PTQ) into DM acceleration. However, the output distributions of noise estimation networks change with time-step, making previous PTQ methods fail in DMs since they are designed for single-time step scenarios. To devise a DM-specific PTQ method, we explore PTQ on DM in three aspects: quantized operations, calibration dataset, and calibration metric. We summarize and use several observations derived from all-inclusive …

Talking head synthesis is a promising approach for the video production industry. Recently, a lot of effort has been devoted in this research area to improve the generation quality or enhance the model generalization. However, there are few works able to address both issues simultaneously, which is essential for practical applications. To this end, in this paper, we turn attention to the emerging powerful Latent Diffusion Models, and model the Talking head generation as an audio-driven temporally coherent denoising process (DiffTalk). More specifically, instead of employing audio signals as the single driving factor, we investigate the control mechanism of the talking face, and incorporate reference face images and landmarks as conditions for personality-aware generalized synthesis. In this way, the proposed DiffTalk is capable of producing high-quality talking head videos in synchronization with the source audio, and more importantly, it can be naturally generalized across different identities without any further fine-tuning. Additionally, our DiffTalk can be gracefully tailored for higher-resolution synthesis with negligible extra computational cost. Extensive experiments show that the proposed DiffTalk efficiently synthesizes high-fidelity audio-driven talking head videos for generalized novel identities. For more video results, please refer to https://sstzal.github.io/DiffTalk/.

Mask-guided matting has shown great practicality compared to traditional trimap-based methods. The mask-guided approach takes an easily-obtainable coarse mask as guidance and produces an accurate alpha matte. To extend the success toward practical usage, we tackle mask-guided matting in the wild, which covers a wide range of categories in their complex context robustly. To this end, we propose a simple yet effective learning framework based on two core insights: 1) learning a generalized matting model that can better understand the given mask guidance and 2) leveraging weak supervision datasets (e.g., instance segmentation dataset) to alleviate the limited diversity and scale of existing matting datasets. Extensive experimental results on multiple benchmarks, consisting of a newly proposed synthetic benchmark (Composition-Wild) and existing natural datasets, demonstrate the superiority of the proposed method. Moreover, we provide appealing results on new practical applications (e.g., panoptic matting and mask-guided video matting), showing the great generality and potential of our model.

Existing autoregressive models follow the two-stage generation paradigm that first learns a codebook in the latent space for image reconstruction and then completes the image generation autoregressively based on the learned codebook. However, existing codebook learning simply models all local region information of images without distinguishing their different perceptual importance, which brings redundancy in the learned codebook that not only limits the next stage’s autoregressive model’s ability to model important structure but also results in high training cost and slow generation speed. In this study, we borrow the idea of importance perception from classical image coding theory and propose a novel two-stage framework, which consists of Masked Quantization VAE (MQ-VAE) and Stackformer, to relieve the model from modeling redundancy. Specifically, MQ-VAE incorporates an adaptive mask module for masking redundant region features before quantization and an adaptive de-mask module for recovering the original grid image feature map to faithfully reconstruct the original images after quantization. Then, Stackformer learns to predict the combination of the next code and its position in the feature map. Comprehensive experiments on various image generation validate our effectiveness and efficiency.

Videos stored on mobile devices or delivered on the Internet are usually in compressed format and are of various unknown compression parameters, but most video super-resolution (VSR) methods often assume ideal inputs resulting in large performance gap between experimental settings and real-world applications. In spite of a few pioneering works being proposed recently to super-resolve the compressed videos, they are not specially designed to deal with videos of various levels of compression. In this paper, we propose a novel and practical compression-aware video super-resolution model, which could adapt its video enhancement process to the estimated compression level. A compression encoder is designed to model compression levels of input frames, and a base VSR model is then conditioned on the implicitly computed representation by inserting compression-aware modules. In addition, we propose to further strengthen the VSR model by taking full advantage of meta data that is embedded naturally in compressed video streams in the procedure of information fusion. Extensive experiments are conducted to demonstrate the effectiveness and efficiency of the proposed method on compressed VSR benchmarks.

Thanks to advances in computer vision and AI, there has been a large growth in the demand for cloud-based visual analytics in which images captured by a low-powered edge device are transmitted to the cloud for analytics. Use of conventional codecs (JPEG, MPEG, HEVC, etc.) for compressing such data introduces artifacts that can seriously degrade the performance of the downstream analytic tasks. Split-DNN computing has emerged as a paradigm to address such usages, in which a DNN is partitioned into a client-side portion and a server side portion. Low-complexity neural networks called ‘bottleneck units’ are introduced at the split point to transform the intermediate layer features into a lower-dimensional representation better suited for compression and transmission. Optimizing the pipeline for both compression and task-performance requires high-quality estimates of the information-theoretic rate of the intermediate features. Most works on compression for image analytics use heuristic approaches to estimate the rate, leading to suboptimal performance. We propose a high-quality ‘neural rate-estimator’ to address this gap. We interpret the lower-dimensional bottleneck output as a latent representation of the intermediate feature and cast the rate-distortion optimization problem as one of training an equivalent variational auto-encoder with an appropriate loss function. We show that this …

Existing implicit neural representation (INR) methods do not fully exploit spatiotemporal redundancies in videos. Index-based INRs ignore the content-specific spatial features and hybrid INRs ignore the contextual dependency on adjacent frames, leading to poor modeling capability for scenes with large motion or dynamics. We analyze this limitation from the perspective of function fitting and reveal the importance of frame difference. To use explicit motion information, we propose Difference Neural Representation for Videos (DNeRV), which consists of two streams for content and frame difference. We also introduce a collaborative content unit for effective feature fusion. We test DNeRV for video compression, inpainting, and interpolation. DNeRV achieves competitive results against the state-of-the-art neural compression approaches and outperforms existing implicit methods on downstream inpainting and interpolation for 960 × 1920 videos.

Implicit neural representations (INR) have gained significant popularity for signal and image representation for many end-tasks, such as superresolution, 3D modeling, and more. Most INR architectures rely on sinusoidal positional encoding, which accounts for high-frequency information in data. However, the finite encoding size restricts the model’s representational power. Higher representational power is needed to go from representing a single given image to representing large and diverse datasets. Our approach addresses this gap by representing an image with a polynomial function and eliminates the need for positional encodings. Therefore, to achieve a progressively higher degree of polynomial representation, we use element-wise multiplications between features and affine-transformed coordinate locations after every ReLU layer. The proposed method is evaluated qualitatively and quantitatively on large datasets like ImageNet. The proposed Poly-INR model performs comparably to state-of-the-art generative models without any convolution, normalization, or self-attention layers, and with far fewer trainable parameters. With much fewer training parameters and higher representative power, our approach paves the way for broader adoption of INR models for generative modeling tasks in complex domains. The code is available at https://github.com/Rajhans0/Poly_INR

Barlow Twins and VICReg are self-supervised representation learning models that use regularizers to decorrelate features. Although these models are as effective as conventional representation learning models, their training can be computationally demanding if the dimension d of the projected embeddings is high. As the regularizers are defined in terms of individual elements of a cross-correlation or covariance matrix, computing the loss for n samples takes O(n d^2) time. In this paper, we propose a relaxed decorrelating regularizer that can be computed in O(n d log d) time by Fast Fourier Transform. We also propose an inexpensive technique to mitigate undesirable local minima that develop with the relaxation. The proposed regularizer exhibits accuracy comparable to that of existing regularizers in downstream tasks, whereas their training requires less memory and is faster for large d. The source code is available.

High-resolution images enable neural networks to learn richer visual representations. However, this improved performance comes at the cost of growing computational complexity, hindering their usage in latency-sensitive applications. As not all pixels are equal, skipping computations for less-important regions offers a simple and effective measure to reduce the computation. This, however, is hard to be translated into actual speedup for CNNs since it breaks the regularity of the dense convolution workload. In this paper, we introduce SparseViT that revisits activation sparsity for recent window-based vision transformers (ViTs). As window attentions are naturally batched over blocks, actual speedup with window activation pruning becomes possible: i.e., ~50% latency reduction with 60% sparsity. Different layers should be assigned with different pruning ratios due to their diverse sensitivities and computational costs. We introduce sparsity-aware adaptation and apply the evolutionary search to efficiently find the optimal layerwise sparsity configuration within the vast search space. SparseViT achieves speedups of 1.5x, 1.4x, and 1.3x compared to its dense counterpart in monocular 3D object detection, 2D instance segmentation, and 2D semantic segmentation, respectively, with negligible to no loss of accuracy.

While some studies have proven that Swin Transformer (Swin) with window self-attention (WSA) is suitable for single image super-resolution (SR), the plain WSA ignores the broad regions when reconstructing high-resolution images due to a limited receptive field. In addition, many deep learning SR methods suffer from intensive computations. To address these problems, we introduce the N-Gram context to the low-level vision with Transformers for the first time. We define N-Gram as neighboring local windows in Swin, which differs from text analysis that views N-Gram as consecutive characters or words. N-Grams interact with each other by sliding-WSA, expanding the regions seen to restore degraded pixels. Using the N-Gram context, we propose NGswin, an efficient SR network with SCDP bottleneck taking multi-scale outputs of the hierarchical encoder. Experimental results show that NGswin achieves competitive performance while maintaining an efficient structure when compared with previous leading methods. Moreover, we also improve other Swin-based SR methods with the N-Gram context, thereby building an enhanced model: SwinIR-NG. Our improved SwinIR-NG outperforms the current best lightweight SR approaches and establishes state-of-the-art results. Codes are available at https://github.com/rami0205/NGramSwin.

Self-attention mechanism has been a key factor in the recent progress of Vision Transformer (ViT), which enables adaptive feature extraction from global contexts. However, existing self-attention methods either adopt sparse global attention or window attention to reduce the computation complexity, which may compromise the local feature learning or subject to some handcrafted designs. In contrast, local attention, which restricts the receptive field of each query to its own neighboring pixels, enjoys the benefits of both convolution and self-attention, namely local inductive bias and dynamic feature selection. Nevertheless, current local attention modules either use inefficient Im2Col function or rely on specific CUDA kernels that are hard to generalize to devices without CUDA support. In this paper, we propose a novel local attention module, Slide Attention, which leverages common convolution operations to achieve high efficiency, flexibility and generalizability. Specifically, we first re-interpret the column-based Im2Col function from a new row-based perspective and use Depthwise Convolution as an efficient substitution. On this basis, we propose a deformed shifting module based on the re-parameterization technique, which further relaxes the fixed key/value positions to deformed features in the local region. In this way, our module realizes the local attention paradigm in both efficient and flexible …

Although vision transformers (ViTs) have shown promising results in various computer vision tasks recently, their high computational cost limits their practical applications. Previous approaches that prune redundant tokens have demonstrated a good trade-off between performance and computation costs. Nevertheless, errors caused by pruning strategies can lead to significant information loss. Our quantitative experiments reveal that the impact of pruned tokens on performance should be noticeable. To address this issue, we propose a novel joint Token Pruning & Squeezing module (TPS) for compressing vision transformers with higher efficiency. Firstly, TPS adopts pruning to get the reserved and pruned subsets. Secondly, TPS squeezes the information of pruned tokens into partial reserved tokens via the unidirectional nearest-neighbor matching and similarity-oriented fusing steps. Compared to state-of-the-art methods, our approach outperforms them under all token pruning intensities. Especially while shrinking DeiT-tiny&small computational budgets to 35%, it improves the accuracy by 1%-6% compared with baselines on ImageNet classification. The proposed method can accelerate the throughput of DeiT-small beyond DeiT-tiny, while its accuracy surpasses DeiT-tiny by 4.78%. Experiments on various transformers demonstrate the effectiveness of our method, while analysis experiments prove our higher robustness to the errors of the token pruning policy. Code is available at https://github.com/megvii-research/TPS-CVPR2023.

Current attention algorithms (e.g., self-attention) are stimulus-driven and highlight all the salient objects in an image. However, intelligent agents like humans often guide their attention based on the high-level task at hand, focusing only on task-related objects. This ability of task-guided top-down attention provides task-adaptive representation and helps the model generalize to various tasks. In this paper, we consider top-down attention from a classic Analysis-by-Synthesis (AbS) perspective of vision. Prior work indicates a functional equivalence between visual attention and sparse reconstruction; we show that an AbS visual system that optimizes a similar sparse reconstruction objective modulated by a goal-directed top-down signal naturally simulates top-down attention. We further propose Analysis-by-Synthesis Vision Transformer (AbSViT), which is a top-down modulated ViT model that variationally approximates AbS, and achieves controllable top-down attention. For real-world applications, AbSViT consistently improves over baselines on Vision-Language tasks such as VQA and zero-shot retrieval where language guides the top-down attention. AbSViT can also serve as a general backbone, improving performance on classification, semantic segmentation, and model robustness. Project page: https://sites.google.com/view/absvit.

Deep artificial neural networks (DNNs) trained through backpropagation provide effective models of the mammalian visual system, accurately capturing the hierarchy of neural responses through primary visual cortex to inferior temporal cortex (IT). However, the ability of these networks to explain representations in higher cortical areas is relatively lacking and considerably less well researched. For example, DNNs have been less successful as a model of the egocentric to allocentric transformation embodied by circuits in retrosplenial and posterior parietal cortex. We describe a novel scene perception benchmark inspired by a hippocampal dependent task, designed to probe the ability of DNNs to transform scenes viewed from different egocentric perspectives. Using a network architecture inspired by the connectivity between temporal lobe structures and the hippocampus, we demonstrate that DNNs trained using a triplet loss can learn this task. Moreover, by enforcing a factorized latent space, we can split information propagation into “what” and “where” pathways, which we use to reconstruct the input. This allows us to beat the state-of-the-art for unsupervised object segmentation on the CATER and MOVi-A,B,C benchmarks.

Masked Image Modeling (MIM) achieves outstanding success in self-supervised representation learning. Unfortunately, MIM models typically have huge computational burden and slow learning process, which is an inevitable obstacle for their industrial applications. Although the lower layers play the key role in MIM, existing MIM models conduct reconstruction task only at the top layer of encoder. The lower layers are not explicitly guided and the interaction among their patches is only used for calculating new activations. Considering the reconstruction task requires non-trivial inter-patch interactions to reason target signals, we apply it to multiple local layers including lower and upper layers. Further, since the multiple layers expect to learn the information of different scales, we design local multi-scale reconstruction, where the lower and upper layers reconstruct fine-scale and coarse-scale supervision signals respectively. This design not only accelerates the representation learning process by explicitly guiding multiple layers, but also facilitates multi-scale semantical understanding to the input. Extensive experiments show that with significantly less pre-training burden, our model achieves comparable or better performance on classification, detection and segmentation tasks than existing MIM models.

Self-supervised learning (SSL) has delivered superior performance on a variety of downstream vision tasks. Two main-stream SSL frameworks have been proposed, i.e., Instance Discrimination (ID) and Masked Image Modeling (MIM). ID pulls together representations from different views of the same image, while avoiding feature collapse. It lacks spatial sensitivity, which requires modeling the local structure within each image. On the other hand, MIM reconstructs the original content given a masked image. It instead does not have good semantic alignment, which requires projecting semantically similar views into nearby representations. To address this dilemma, we observe that (1) semantic alignment can be achieved by matching different image views with strong augmentations; (2) spatial sensitivity can benefit from predicting dense representations with masked images. Driven by these analysis, we propose Siamese Image Modeling (SiameseIM), which predicts the dense representations of an augmented view, based on another masked view from the same image but with different augmentations. SiameseIM uses a Siamese network with two branches. The online branch encodes the first view, and predicts the second view’s representation according to the relative positions between these two views. The target branch produces the target by encoding the second view. SiameseIM can surpass both ID and …

Generative modeling and representation learning are two key tasks in computer vision. However, these models are typically trained independently, which ignores the potential for each task to help the other, and leads to training and model maintenance overheads. In this work, we propose MAsked Generative Encoder (MAGE), the first framework to unify SOTA image generation and self-supervised representation learning. Our key insight is that using variable masking ratios in masked image modeling pre-training can allow generative training (very high masking ratio) and representation learning (lower masking ratio) under the same training framework. Inspired by previous generative models, MAGE uses semantic tokens learned by a vector-quantized GAN at inputs and outputs, combining this with masking. We can further improve the representation by adding a contrastive loss to the encoder output. We extensively evaluate the generation and representation learning capabilities of MAGE. On ImageNet-1K, a single MAGE ViT-L model obtains 9.10 FID in the task of class-unconditional image generation and 78.9% top-1 accuracy for linear probing, achieving state-of-the-art performance in both image generation and representation learning. Code is available at https://github.com/LTH14/mage.

For the visible-infrared person re-identification (VIReID) task, one of the major challenges is the modality gaps between visible (VIS) and infrared (IR) images. However, the training samples are usually limited, while the modality gaps are too large, which leads that the existing methods cannot effectively mine diverse cross-modality clues. To handle this limitation, we propose a novel augmentation network in the embedding space, called diverse embedding expansion network (DEEN). The proposed DEEN can effectively generate diverse embeddings to learn the informative feature representations and reduce the modality discrepancy between the VIS and IR images. Moreover, the VIReID model may be seriously affected by drastic illumination changes, while all the existing VIReID datasets are captured under sufficient illumination without significant light changes. Thus, we provide a low-light cross-modality (LLCM) dataset, which contains 46,767 bounding boxes of 1,064 identities captured by 9 RGB/IR cameras. Extensive experiments on the SYSU-MM01, RegDB and LLCM datasets show the superiority of the proposed DEEN over several other state-of-the-art methods. The code and dataset are released at: https://github.com/ZYK100/LLCM

In the field of human pose estimation, regression-based methods have been dominated in terms of speed, while heatmap-based methods are far ahead in terms of performance. How to take advantage of both schemes remains a challenging problem. In this paper, we propose a novel human pose estimation framework termed DistilPose, which bridges the gaps between heatmap-based and regression-based methods. Specifically, DistilPose maximizes the transfer of knowledge from the teacher model (heatmap-based) to the student model (regression-based) through Token-distilling Encoder (TDE) and Simulated Heatmaps. TDE aligns the feature spaces of heatmap-based and regression-based models by introducing tokenization, while Simulated Heatmaps transfer explicit guidance (distribution and confidence) from teacher heatmaps into student models. Extensive experiments show that the proposed DistilPose can significantly improve the performance of the regression-based models while maintaining efficiency. Specifically, on the MSCOCO validation dataset, DistilPose-S obtains 71.6% mAP with 5.36M parameter, 2.38 GFLOPs and 40.2 FPS, which saves 12.95x, 7.16x computational cost and is 4.9x faster than its teacher model with only 0.9 points performance drop. Furthermore, DistilPose-L obtains 74.4% mAP on MSCOCO validation dataset, achieving a new state-of-the-art among predominant regression-based models.

We present a novel graph Transformer generative adversarial network (GTGAN) to learn effective graph node relations in an end-to-end fashion for the challenging graph-constrained house generation task. The proposed graph-Transformer-based generator includes a novel graph Transformer encoder that combines graph convolutions and self-attentions in a Transformer to model both local and global interactions across connected and non-connected graph nodes. Specifically, the proposed connected node attention (CNA) and non-connected node attention (NNA) aim to capture the global relations across connected nodes and non-connected nodes in the input graph, respectively. The proposed graph modeling block (GMB) aims to exploit local vertex interactions based on a house layout topology. Moreover, we propose a new node classification-based discriminator to preserve the high-level semantic and discriminative node features for different house components. Finally, we propose a novel graph-based cycle-consistency loss that aims at maintaining the relative spatial relationships between ground truth and predicted graphs. Experiments on two challenging graph-constrained house generation tasks (i.e., house layout and roof generation) with two public datasets demonstrate the effectiveness of GTGAN in terms of objective quantitative scores and subjective visual realism. New state-of-the-art results are established by large margins on both tasks.

Wires and powerlines are common visual distractions that often undermine the aesthetics of photographs. The manual process of precisely segmenting and removing them is extremely tedious and may take up to hours, especially on high-resolution photos where wires may span the entire space. In this paper, we present an automatic wire clean-up system that eases the process of wire segmentation and removal/inpainting to within a few seconds. We observe several unique challenges: wires are thin, lengthy, and sparse. These are rare properties of subjects that common segmentation tasks cannot handle, especially in high-resolution images. We thus propose a two-stage method that leverages both global and local context to accurately segment wires in high-resolution images efficiently, and a tile-based inpainting strategy to remove the wires given our predicted segmentation masks. We also introduce the first wire segmentation benchmark dataset, WireSegHR. Finally, we demonstrate quantitatively and qualitatively that our wire clean-up system enables fully automated wire removal for great generalization to various wire appearances.

We present a novel approach to perform instance segmentation, and counting, for densely packed self-similar trees using a top-view RGB image sequence. We propose a solution that leverages pixel content, shape, and self-occlusion. First, we perform an initial over-segmentation of the image sequence and aggregate structural characteristics into a contour graph with temporal information incorporated. Second, using a graph convolutional network and its inherent local messaging passing abilities, we merge adjacent tree crown patches into a final set of tree crowns. Per various studies and comparisons, our method is superior to all prior methods and results in high-accuracy instance segmentation and counting, despite the trees being tightly packed. Finally, we provide various forest image sequence datasets suitable for subsequent benchmarking and evaluation captured at different altitudes and leaf conditions.

In this paper, we efficiently transfer the surpassing representation power of the vision foundation models, such as ViT and Swin, for video understanding with only a few trainable parameters. Previous adaptation methods have simultaneously considered spatial and temporal modeling with a unified learnable module but still suffered from fully leveraging the representative capabilities of image transformers. We argue that the popular dual-path (two-stream) architecture in video models can mitigate this problem. We propose a novel DUALPATH adaptation separated into spatial and temporal adaptation paths, where a lightweight bottleneck adapter is employed in each transformer block. Especially for temporal dynamic modeling, we incorporate consecutive frames into a grid-like frameset to precisely imitate vision transformers’ capability that extrapolates relationships between tokens. In addition, we extensively investigate the multiple baselines from a unified perspective in video understanding and compare them with DUALPATH. Experimental results on four action recognition benchmarks prove that pretrained image transformers with DUALPATH can be effectively generalized beyond the data domain.

We present a simple approach which can turn a ViT encoder into an efficient video model, which can seamlessly work with both image and video inputs. By sparsely sampling the inputs, the model is able to do training and inference from both inputs. The model is easily scalable and can be adapted to large-scale pre-trained ViTs without requiring full finetuning. The model achieves SOTA results.

A meaningful video is semantically coherent and changes smoothly. However, most existing fine-grained video representation learning methods learn frame-wise features by aligning frames across videos or exploring relevance between multiple views, neglecting the inherent dynamic process of each video. In this paper, we propose to learn video representations by modeling Video as Stochastic Processes (VSP) via a novel process-based contrastive learning framework, which aims to discriminate between video processes and simultaneously capture the temporal dynamics in the processes. Specifically, we enforce the embeddings of the frame sequence of interest to approximate a goal-oriented stochastic process, i.e., Brownian bridge, in the latent space via a process-based contrastive loss. To construct the Brownian bridge, we adapt specialized sampling strategies under different annotations for both self-supervised and weakly-supervised learning. Experimental results on four datasets show that VSP stands as a state-of-the-art method for various video understanding tasks, including phase progression, phase classification and frame retrieval. Code is available at ‘https://github.com/hengRUC/VSP’.

How to learn discriminative video representation from unlabeled videos is challenging but crucial for video analysis. The latest attempts seek to learn a representation model by predicting the appearance contents in the masked regions. However, simply masking and recovering appearance contents may not be sufficient to model temporal clues as the appearance contents can be easily reconstructed from a single frame. To overcome this limitation, we present Masked Motion Encoding (MME), a new pre-training paradigm that reconstructs both appearance and motion information to explore temporal clues. In MME, we focus on addressing two critical challenges to improve the representation performance: 1) how to well represent the possible long-term motion across multiple frames; and 2) how to obtain fine-grained temporal clues from sparsely sampled videos. Motivated by the fact that human is able to recognize an action by tracking objects’ position changes and shape changes, we propose to reconstruct a motion trajectory that represents these two kinds of change in the masked regions. Besides, given the sparse video input, we enforce the model to reconstruct dense motion trajectories in both spatial and temporal dimensions. Pre-trained with our MME paradigm, the model is able to anticipate long-term and fine-grained motion details. Code …

Current top-leading solutions for video object segmentation (VOS) typically follow a matching-based regime: for each query frame, the segmentation mask is inferred according to its correspondence to previously processed and the first annotated frames. They simply exploit the supervisory signals from the groundtruth masks for learning mask prediction only, without posing any constraint on the space-time correspondence matching, which, however, is the fundamental building block of such regime. To alleviate this crucial yet commonly ignored issue, we devise a correspondence-aware training framework, which boosts matching-based VOS solutions by explicitly encouraging robust correspondence matching during network learning. Through comprehensively exploring the intrinsic coherence in videos on pixel and object levels, our algorithm reinforces the standard, fully supervised training of mask segmentation with label-free, contrastive correspondence learning. Without neither requiring extra annotation cost during training, nor causing speed delay during deployment, nor incurring architectural modification, our algorithm provides solid performance gains on four widely used benchmarks, i.e., DAVIS2016&2017, and YouTube-VOS2018&2019, on the top of famous matching-based VOS solutions. Our implementation will be released.

Previous works on video object segmentation (VOS) are trained on densely annotated videos. Nevertheless, acquiring annotations in pixel level is expensive and time-consuming. In this work, we demonstrate the feasibility of training a satisfactory VOS model on sparsely annotated videos--we merely require two labeled frames per training video while the performance is sustained. We term this novel training paradigm as two-shot video object segmentation, or two-shot VOS for short. The underlying idea is to generate pseudo labels for unlabeled frames during training and to optimize the model on the combination of labeled and pseudo-labeled data. Our approach is extremely simple and can be applied to a majority of existing frameworks. We first pre-train a VOS model on sparsely annotated videos in a semi-supervised manner, with the first frame always being a labeled one. Then, we adopt the pre-trained VOS model to generate pseudo labels for all unlabeled frames, which are subsequently stored in a pseudo-label bank. Finally, we retrain a VOS model on both labeled and pseudo-labeled data without any restrictions on the first frame. For the first time, we present a general way to train VOS models on two-shot VOS datasets. By using 7.3% and 2.9% labeled data of …

Exploring dense matching between the current frame and past frames for long-range context modeling, memory-based methods have demonstrated impressive results in video object segmentation (VOS) recently. Nevertheless, due to the lack of instance understanding ability, the above approaches are oftentimes brittle to large appearance variations or viewpoint changes resulted from the movement of objects and cameras. In this paper, we argue that instance understanding matters in VOS, and integrating it with memory-based matching can enjoy the synergy, which is intuitively sensible from the definition of VOS task, i.e., identifying and segmenting object instances within the video. Towards this goal, we present a two-branch network for VOS, where the query-based instance segmentation (IS) branch delves into the instance details of the current frame and the VOS branch performs spatial-temporal matching with the memory bank. We employ the well-learned object queries from IS branch to inject instance-specific information into the query key, with which the instance-augmented matching is further performed. In addition, we introduce a multi-path fusion block to effectively combine the memory readout with multi-scale features from the instance segmentation decoder, which incorporates high-resolution instance-aware features to produce final segmentation results. Our method achieves state-of-the-art performance on DAVIS 2016/2017 val (92.6% …

In low-level video analyses, effective representations are important to derive the correspondences between video frames. These representations have been learned in a self-supervised fashion from unlabeled images/videos, using carefully designed pretext tasks in some recent studies. However, the previous work concentrates on either spatial-discriminative features or temporal-repetitive features, with little attention to the synergy between spatial and temporal cues. To address this issue, we propose a novel spatial-then-temporal self-supervised learning method. Specifically, we firstly extract spatial features from unlabeled images via contrastive learning, and secondly enhance the features by exploiting the temporal cues in unlabeled videos via reconstructive learning. In the second step, we design a global correlation distillation loss to ensure the learning not to forget the spatial cues, and we design a local correlation distillation loss to combat the temporal discontinuity that harms the reconstruction. The proposed method outperforms the state-of-the-art self-supervised methods, as established by the experimental results on a series of correspondence-based video analysis tasks. Also, we performed ablation studies to verify the effectiveness of the two-step design as well as the distillation losses.

Interpreting visual relationships is a core aspect of comprehensive video understanding. Given a query visual relationship as and a test video, our objective is to localize the subject and object that are connected via the predicate. Given modern visio-lingual understanding capabilities, solving this problem is achievable, provided that there are large-scale annotated training examples available. However, annotating for every combination of subject, object, and predicate is cumbersome, expensive, and possibly infeasible. Therefore, there is a need for models that can learn to spatially and temporally localize subjects and objects that are connected via an unseen predicate using only a few support set videos sharing the common predicate. We address this challenging problem, referred to as few-shot referring relationships in videos for the first time. To this end, we pose the problem as a minimization of an objective function defined over a T-partite random field. Here, the vertices of the random field correspond to candidate bounding boxes for the subject and object, and T represents the number of frames in the test video. This objective function is composed of frame level and visual relationship similarity potentials. To learn these potentials, we use a relation network that takes query-conditioned translational relationship embedding …

Vision transformers (ViTs) have achieved impressive results on various computer vision tasks in the last several years. In this work, we study the capability of frozen ViTs, pretrained only on visual data, to generalize to audio-visual data without finetuning any of its original parameters. To do so, we propose a latent audio-visual hybrid (LAVISH) adapter that adapts pretrained ViTs to audio-visual tasks by injecting a small number of trainable parameters into every layer of a frozen ViT. To efficiently fuse visual and audio cues, our LAVISH adapter uses a small set of latent tokens, which form an attention bottleneck, thus, eliminating the quadratic cost of standard cross-attention. Compared to the existing modality-specific audio-visual methods, our approach achieves competitive or even better performance on various audio-visual tasks while using fewer tunable parameters and without relying on costly audio pretraining or external audio encoders. Our code is available at https://genjib.github.io/project_page/LAVISH/

Different video understanding tasks are typically treated in isolation, and even with distinct types of curated data (e.g., classifying sports in one dataset, tracking animals in another). However, in wearable cameras, the immersive egocentric perspective of a person engaging with the world around them presents an interconnected web of video understanding tasks---hand-object manipulations, navigation in the space, or human-human interactions---that unfold continuously, driven by the person’s goals. We argue that this calls for a much more unified approach. We propose EgoTask Translation (EgoT2), which takes a collection of models optimized on separate tasks and learns to translate their outputs for improved performance on any or all of them at once. Unlike traditional transfer or multi-task learning, EgoT2’s “flipped design” entails separate task-specific backbones and a task translator shared across all tasks, which captures synergies between even heterogeneous tasks and mitigates task competition. Demonstrating our model on a wide array of video tasks from Ego4D, we show its advantages over existing transfer paradigms and achieve top-ranked results on four of the Ego4D 2022 benchmark challenges.

Speech-driven gesture generation is highly challenging due to the random jitters of human motion. In addition, there is an inherent asynchronous relationship between human speech and gestures. To tackle these challenges, we introduce a novel quantization-based and phase-guided motion matching framework. Specifically, we first present a gesture VQ-VAE module to learn a codebook to summarize meaningful gesture units. With each code representing a unique gesture, random jittering problems are alleviated effectively. We then use Levenshtein distance to align diverse gestures with different speech. Levenshtein distance based on audio quantization as a similarity metric of corresponding speech of gestures helps match more appropriate gestures with speech, and solves the alignment problem of speech and gestures well. Moreover, we introduce phase to guide the optimal gesture matching based on the semantics of context or rhythm of audio. Phase guides when text-based or speech-based gestures should be performed to make the generated gestures more natural. Extensive experiments show that our method outperforms recent approaches on speech-driven gesture generation. Our code, database, pre-trained models and demos are available at https://github.com/YoungSeng/QPGesture.

There is a growing demand of automatically synthesizing co-speech gestures for virtual characters. However, it remains a challenge due to the complex relationship between input speeches and target gestures. Most existing works focus on predicting the next gesture that fits the data best, however, such methods are myopic and lack the ability to plan for future gestures. In this paper, we propose a novel reinforcement learning (RL) framework called RACER to generate sequences of gestures that maximize the overall satisfactory. RACER employs a vector quantized variational autoencoder to learn compact representations of gestures and a GPT-based policy architecture to generate coherent sequence of gestures autoregressively. In particular, we propose a contrastive pre-training approach to calculate the rewards, which integrates contextual information into action evaluation and successfully captures the complex relationships between multi-modal speech-gesture data. Experimental results show that our method significantly outperforms existing baselines in terms of both objective metrics and subjective human judgements. Demos can be found at https://github.com/RLracer/RACER.git.

Semi-Supervised Learning can be more beneficial for the video domain compared to images because of its higher annotation cost and dimensionality. Besides, any video understanding task requires reasoning over both spatial and temporal dimensions. In order to learn both the static and motion related features for the semi-supervised action recognition task, existing methods rely on hard input inductive biases like using two-modalities (RGB and Optical-flow) or two-stream of different playback rates. Instead of utilizing unlabeled videos through diverse input streams, we rely on self-supervised video representations, particularly, we utilize temporally-invariant and temporally-distinctive representations. We observe that these representations complement each other depending on the nature of the action. Based on this observation, we propose a student-teacher semi-supervised learning framework, TimeBalance, where we distill the knowledge from a temporally-invariant and a temporally-distinctive teacher. Depending on the nature of the unlabeled video, we dynamically combine the knowledge of these two teachers based on a novel temporal similarity-based reweighting scheme. Our method achieves state-of-the-art performance on three action recognition benchmarks: UCF101, HMDB51, and Kinetics400. Code: https://github.com/DAVEISHAN/TimeBalance.

Current state-of-the-art video models process a video clip as a long sequence of spatio-temporal tokens. However, they do not explicitly model objects, their interactions across the video, and instead process all the tokens in the video. In this paper, we investigate how we can use knowledge of objects to design better video models, namely to process fewer tokens and to improve recognition accuracy. This is in contrast to prior works which either drop tokens at the cost of accuracy, or increase accuracy whilst also increasing the computation required. First, we propose an object-guided token sampling strategy that enables us to retain a small fraction of the input tokens with minimal impact on accuracy. And second, we propose an object-aware attention module that enriches our feature representation with object information and improves overall accuracy. Our resulting framework achieves better performance when using fewer tokens than strong baselines. In particular, we match our baseline with 30%, 40%, and 60% of the input tokens on SomethingElse, Something-something v2, and Epic-Kitchens, respectively. When we use our model to process the same number of tokens as our baseline, we improve by 0.6 to 4.2 points on these datasets.

The self-supervised pretraining paradigm has achieved great success in skeleton-based action recognition. However, these methods treat the motion and static parts equally, and lack an adaptive design for different parts, which has a negative impact on the accuracy of action recognition. To realize the adaptive action modeling of both parts, we propose an Actionlet-Dependent Contrastive Learning method (ActCLR). The actionlet, defined as the discriminative subset of the human skeleton, effectively decomposes motion regions for better action modeling. In detail, by contrasting with the static anchor without motion, we extract the motion region of the skeleton data, which serves as the actionlet, in an unsupervised manner. Then, centering on actionlet, a motion-adaptive data transformation method is built. Different data transformations are applied to actionlet and non-actionlet regions to introduce more diversity while maintaining their own characteristics. Meanwhile, we propose a semantic-aware feature pooling method to build feature representations among motion and static regions in a distinguished manner. Extensive experiments on NTU RGB+D and PKUMMD show that the proposed method achieves remarkable action recognition performance. More visualization and quantitative experiments demonstrate the effectiveness of our method.

Temporal action detection aims to predict the time intervals and the classes of action instances in the video. Despite the promising performance, existing two-stream models exhibit slow inference speed due to their reliance on computationally expensive optical flow. In this paper, we introduce a decomposed cross-modal distillation framework to build a strong RGB-based detector by transferring knowledge of the motion modality. Specifically, instead of direct distillation, we propose to separately learn RGB and motion representations, which are in turn combined to perform action localization. The dual-branch design and the asymmetric training objectives enable effective motion knowledge transfer while preserving RGB information intact. In addition, we introduce a local attentive fusion to better exploit the multimodal complementarity. It is designed to preserve the local discriminability of the features that is important for action localization. Extensive experiments on the benchmarks verify the effectiveness of the proposed method in enhancing RGB-based action detectors. Notably, our framework is agnostic to backbones and detection heads, bringing consistent gains across different model combinations.

Most state-of-the-art methods for action segmentation are based on single input modalities or naïve fusion of multiple data sources. However, effective fusion of complementary information can potentially strengthen segmentation models and make them more robust to sensor noise and more accurate with smaller training datasets. In order to improve multimodal representation learning for action segmentation, we propose to disentangle hidden features of a multi-stream segmentation model into modality-shared components, containing common information across data sources, and private components; we then use an attention bottleneck to capture long-range temporal dependencies in the data while preserving disentanglement in consecutive processing layers. Evaluation on 50salads, Breakfast and RARP45 datasets shows that our multimodal approach outperforms different data fusion baselines on both multiview and multimodal data sources, obtaining competitive or better results compared with the state-of-the-art. Our model is also more robust to additive sensor noise and can achieve performance on par with strong video baselines even with less training data.

Weakly-supervised temporal action localization aims to localize and recognize actions in untrimmed videos with only video-level category labels during training. Without instance-level annotations, most existing methods follow the Segment-based Multiple Instance Learning (S-MIL) framework, where the predictions of segments are supervised by the labels of videos. However, the objective for acquiring segment-level scores during training is not consistent with the target for acquiring proposal-level scores during testing, leading to suboptimal results. To deal with this problem, we propose a novel Proposal-based Multiple Instance Learning (P-MIL) framework that directly classifies the candidate proposals in both the training and testing stages, which includes three key designs: 1) a surrounding contrastive feature extraction module to suppress the discriminative short proposals by considering the surrounding contrastive information, 2) a proposal completeness evaluation module to inhibit the low-quality proposals with the guidance of the completeness pseudo labels, and 3) an instance-level rank consistency loss to achieve robust detection by leveraging the complementarity of RGB and FLOW modalities. Extensive experimental results on two challenging benchmarks including THUMOS14 and ActivityNet demonstrate the superior performance of our method. Our code is available at github.com/OpenSpaceAI/CVPR2023_P-MIL.

Action quality assessment (AQA) has become an emerging topic since it can be extensively applied in numerous scenarios. However, most existing methods and datasets focus on single-person short-sequence scenes, hindering the application of AQA in more complex situations. To address this issue, we construct a new multi-person long-form video dataset for action quality assessment named LOGO. Distinguished in scenario complexity, our dataset contains 200 videos from 26 artistic swimming events with 8 athletes in each sample along with an average duration of 204.2 seconds. As for richness in annotations, LOGO includes formation labels to depict group information of multiple athletes and detailed annotations on action procedures. Furthermore, we propose a simple yet effective method to model relations among athletes and reason about the potential temporal logic in long-form videos. Specifically, we design a group-aware attention module, which can be easily plugged into existing AQA methods, to enrich the clip-wise representations based on contextual group information. To benchmark LOGO, we systematically conduct investigations on the performance of several popular methods in AQA and action segmentation. The results reveal the challenges our dataset brings. Extensive experiments also show that our approach achieves state-of-the-art on the LOGO dataset. The dataset and code will …

This paper presents an investigation into long-tail video recognition. We demonstrate that, unlike naturally-collected video datasets and existing long-tail image benchmarks, current video benchmarks fall short on multiple long-tailed properties. Most critically, they lack few-shot classes in their tails. In response, we propose new video benchmarks that better assess long-tail recognition, by sampling subsets from two datasets: SSv2 and VideoLT. We then propose a method, Long-Tail Mixed Reconstruction (LMR), which reduces overfitting to instances from few-shot classes by reconstructing them as weighted combinations of samples from head classes. LMR then employs label mixing to learn robust decision boundaries. It achieves state-of-the-art average class accuracy on EPIC-KITCHENS and the proposed SSv2-LT and VideoLT-LT. Benchmarks and code at: github.com/tobyperrett/lmr

The sound effects that designers add to videos are designed to convey a particular artistic effect and, thus, may be quite different from a scene’s true sound. Inspired by the challenges of creating a soundtrack for a video that differs from its true sound, but that nonetheless matches the actions occurring on screen, we propose the problem of conditional Foley. We present the following contributions to address this problem. First, we propose a pretext task for training our model to predict sound for an input video clip using a conditional audio-visual clip sampled from another time within the same source video. Second, we propose a model for generating a soundtrack for a silent input video, given a user-supplied example that specifies what the video should “sound like”. We show through human studies and automated evaluation metrics that our model successfully generates sound from video, while varying its output according to the content of a supplied example.

Sequential video understanding, as an emerging video understanding task, has driven lots of researchers’ attention because of its goal-oriented nature. This paper studies weakly supervised sequential video understanding where the accurate time-stamp level text-video alignment is not provided. We solve this task by borrowing ideas from CLIP. Specifically, we use a transformer to aggregate frame-level features for video representation and use a pre-trained text encoder to encode the texts corresponding to each action and the whole video, respectively. To model the correspondence between text and video, we propose a multiple granularity loss, where the video-paragraph contrastive loss enforces matching between the whole video and the complete script, and a fine-grained frame-sentence contrastive loss enforces the matching between each action and its description. As the frame-sentence correspondence is not available, we propose to use the fact that video actions happen sequentially in the temporal domain to generate pseudo frame-sentence correspondence and supervise the network training with the pseudo labels. Extensive experiments on video sequence verification and text-to-video matching show that our method outperforms baselines by a large margin, which validates the effectiveness of our proposed approach. Code is available at https://github.com/svip-lab/WeakSVR.

Given an untrimmed video, temporal sentence grounding (TSG) aims to locate a target moment semantically according to a sentence query. Although previous respectable works have made decent success, they only focus on high-level visual features extracted from the consecutive decoded frames and fail to handle the compressed videos for query modelling, suffering from insufficient representation capability and significant computational complexity during training and testing. In this paper, we pose a new setting, compressed-domain TSG, which directly utilizes compressed videos rather than fully-decompressed frames as the visual input. To handle the raw video bit-stream input, we propose a novel Three-branch Compressed-domain Spatial-temporal Fusion (TCSF) framework, which extracts and aggregates three kinds of low-level visual features (I-frame, motion vector and residual features) for effective and efficient grounding. Particularly, instead of encoding the whole decoded frames like previous works, we capture the appearance representation by only learning the I-frame feature to reduce delay or latency. Besides, we explore the motion information not only by learning the motion vector feature, but also by exploring the relations of neighboring frames via the residual feature. In this way, a three-branch spatial-temporal attention layer with an adaptive motion-appearance fusion module is further designed to extract and aggregate …

We propose Video Localized Narratives, a new form of multimodal video annotations connecting vision and language. In the original Localized Narratives, annotators speak and move their mouse simultaneously on an image, thus grounding each word with a mouse trace segment. However, this is challenging on a video. Our new protocol empowers annotators to tell the story of a video with Localized Narratives, capturing even complex events involving multiple actors interacting with each other and with several passive objects. We annotated 20k videos of the OVIS, UVO, and Oops datasets, totalling 1.7M words. Based on this data, we also construct new benchmarks for the video narrative grounding and video question answering tasks, and provide reference results from strong baseline models. Our annotations are available at https://google.github.io/video-localized-narratives/.

Contrastive learning-based video-language representation learning approaches, e.g., CLIP, have achieved outstanding performance, which pursue semantic interaction upon pre-defined video-text pairs. To clarify this coarse-grained global interaction and move a step further, we have to encounter challenging shell-breaking interactions for fine-grained cross-modal learning. In this paper, we creatively model video-text as game players with multivariate cooperative game theory to wisely handle the uncertainty during fine-grained semantic interaction with diverse granularity, flexible combination, and vague intensity. Concretely, we propose Hierarchical Banzhaf Interaction (HBI) to value possible correspondence between video frames and text words for sensitive and explainable cross-modal contrast. To efficiently realize the cooperative game of multiple video frames and multiple text words, the proposed method clusters the original video frames (text words) and computes the Banzhaf Interaction between the merged tokens. By stacking token merge modules, we achieve cooperative games at different semantic levels. Extensive experiments on commonly used text-video retrieval and video-question answering benchmarks with superior performances justify the efficacy of our HBI. More encouragingly, it can also serve as a visualization tool to promote the understanding of cross-modal interaction, which may have a far-reaching impact on the community. Project page is available at https://jpthu17.github.io/HBI/.

Multimodal alignment facilitates the retrieval of instances from one modality when queried using another. In this paper, we consider a novel setting where such an alignment is between (i) instruction steps that are depicted as assembly diagrams (commonly seen in Ikea assembly manuals) and (ii) video segments from in-the-wild videos; these videos comprising an enactment of the assembly actions in the real world. To learn this alignment, we introduce a novel supervised contrastive learning method that learns to align videos with the subtle details in the assembly diagrams, guided by a set of novel losses. To study this problem and demonstrate the effectiveness of our method, we introduce a novel dataset: IAW---for Ikea assembly in the wild---consisting of 183 hours of videos from diverse furniture assembly collections and nearly 8,300 illustrations from their associated instruction manuals and annotated for their ground truth alignments. We define two tasks on this dataset: First, nearest neighbor retrieval between video segments and illustrations, and, second, alignment of instruction steps and the segments for each video. Extensive experiments on IAW demonstrate superior performances of our approach against alternatives.

There has been a recent explosion of impressive generative models that can produce high quality images (or videos) conditioned on text descriptions. However, all such approaches rely on conditional sentences that contain unambiguous descriptions of scenes and main actors in them. Therefore employing such models for more complex task of story visualization, where naturally references and co-references exist, and one requires to reason about when to maintain consistency of actors and backgrounds across frames/scenes, and when not to, based on story progression, remains a challenge. In this work, we address the aforementioned challenges and propose a novel autoregressive diffusion-based framework with a visual memory module that implicitly captures the actor and background context across the generated frames. Sentence-conditioned soft attention over the memories enables effective reference resolution and learns to maintain scene and actor consistency when needed. To validate the effectiveness of our approach, we extend the MUGEN dataset and introduce additional characters, backgrounds and referencing in multi-sentence storylines. Our experiments for story generation on the MUGEN, the PororoSV and the FlintstonesSV dataset show that our method not only outperforms prior state-of-the-art in generating frames with high visual quality, which are consistent with the story, but also models appropriate correspondences …

Modelling and understanding time remains a challenge in contemporary video understanding models. With language emerging as a key driver towards powerful generalization, it is imperative for foundational video-language models to have a sense of time. In this paper, we consider a specific aspect of temporal understanding: consistency of time order as elicited by before/after relations. We establish that seven existing video-language models struggle to understand even such simple temporal relations. We then question whether it is feasible to equip these foundational models with temporal awareness without re-training them from scratch. Towards this, we propose a temporal adaptation recipe on top of one such model, VideoCLIP, based on post-pretraining on a small amount of video-text data. We conduct a zero-shot evaluation of the adapted models on six datasets for three downstream tasks which require varying degrees of time awareness. We observe encouraging performance gains especially when the task needs higher time awareness. Our work serves as a first step towards probing and instilling a sense of time in existing video-language models without the need for data and compute-intense training from scratch.

Movie story analysis requires understanding characters’ emotions and mental states. Towards this goal, we formulate emotion understanding as predicting a diverse and multi-label set of emotions at the level of a movie scene and for each character. We propose EmoTx, a multimodal Transformer-based architecture that ingests videos, multiple characters, and dialog utterances to make joint predictions. By leveraging annotations from the MovieGraphs dataset, we aim to predict classic emotions (e.g. happy, angry) and other mental states (e.g. honest, helpful). We conduct experiments on the most frequently occurring 10 and 25 labels, and a mapping that clusters 181 labels to 26. Ablation studies and comparison against adapted state-of-the-art emotion recognition approaches shows the effectiveness of EmoTx. Analyzing EmoTx’s self-attention scores reveals that expressive emotions often look at character tokens while other mental states rely on video and dialog cues.

Human body trajectories are a salient cue to identify actions in video. Such body trajectories are mainly conveyed by hands and face across consecutive frames in sign language. However, current methods in continuous sign language recognition(CSLR) usually process frames independently to capture frame-wise features, thus failing to capture cross-frame trajectories to effectively identify a sign. To handle this limitation, we propose correlation network (CorrNet) to explicitly leverage body trajectories across frames to identify signs. In specific, an identification module is first presented to emphasize informative regions in each frame that are beneficial in expressing a sign. A correlation module is then proposed to dynamically compute correlation maps between current frame and adjacent neighbors to capture cross-frame trajectories. As a result, the generated features are able to gain an overview of local temporal movements to identify a sign. Thanks to its special attention on body trajectories, CorrNet achieves new state-of-the-art accuracy on four large-scale datasets, PHOENIX14, PHOENIX14-T, CSL-Daily, and CSL. A comprehensive comparison between CorrNet and previous spatial-temporal reasoning methods verifies its effectiveness. Visualizations are given to demonstrate the effects of CorrNet on emphasizing human body trajectories across adjacent frames.

Sarcasm indicates the literal meaning is contrary to the real attitude. Considering the popularity and complementarity of image-text data, we investigate the task of multi-modal sarcasm detection. Different from other multi-modal tasks, for the sarcastic data, there exists intrinsic incongruity between a pair of image and text as demonstrated in psychological theories. To tackle this issue, we propose a Dual Incongruity Perceiving (DIP) network consisting of two branches to mine the sarcastic information from factual and affective levels. For the factual aspect, we introduce a channel-wise reweighting strategy to obtain semantically discriminative embeddings, and leverage gaussian distribution to model the uncertain correlation caused by the incongruity. The distribution is generated from the latest data stored in the memory bank, which can adaptively model the difference of semantic similarity between sarcastic and non-sarcastic data. For the affective aspect, we utilize siamese layers with shared parameters to learn cross-modal sentiment information. Furthermore, we use the polarity value to construct a relation graph for the mini-batch, which forms the continuous contrastive loss to acquire affective embeddings. Extensive experiments demonstrate that our proposed method performs favorably against state-of-the-art approaches. Our code is released on https://github.com/downdric/MSD.

Most sign language translation (SLT) methods to date require the use of gloss annotations to provide additional supervision information, however, the acquisition of gloss is not easy. To solve this problem, we first perform an analysis of existing models to confirm how gloss annotations make SLT easier. We find that it can provide two aspects of information for the model, 1) it can help the model implicitly learn the location of semantic boundaries in continuous sign language videos, 2) it can help the model understand the sign language video globally. We then propose gloss attention, which enables the model to keep its attention within video segments that have the same semantics locally, just as gloss helps existing models do. Furthermore, we transfer the knowledge of sentence-to-sentence similarity from the natural language model to our gloss attention SLT network (GASLT) to help it understand sign language videos at the sentence level. Experimental results on multiple large-scale sign language datasets show that our proposed GASLT model significantly outperforms existing methods. Our code is provided in https://github.com/YinAoXiong/GASLT.

Object-goal visual navigation aims at steering an agent toward an object via a series of moving steps. Previous works mainly focus on learning informative visual representations for navigation, but overlook the impacts of navigation states on the effectiveness and efficiency of navigation. We observe that high relevance among navigation states will cause navigation inefficiency or failure for existing methods. In this paper, we present a History-inspired Navigation Policy Learning (HiNL) framework to estimate navigation states effectively by exploring relationships among historical navigation states. In HiNL, we propose a History-aware State Estimation (HaSE) module to alleviate the impacts of dominant historical states on the current state estimation. Meanwhile, HaSE also encourages an agent to be alert to the current observation changes, thus enabling the agent to make valid actions. Furthermore, we design a History-based State Regularization (HbSR) to explicitly suppress the correlation among navigation states in training. As a result, our agent can update states more effectively while reducing the correlations among navigation states. Experiments on the artificial platform AI2-THOR (i.e.,, iTHOR and RoboTHOR) demonstrate that HiNL significantly outperforms state-of-the-art methods on both Success Rate and SPL in unseen testing environments.

To be successful, Vision-and-Language Navigation (VLN) agents must be able to ground instructions to actions based on their surroundings. In this work, we develop a methodology to study agent behavior on a skill-specific basis -- examining how well existing agents ground instructions about stopping, turning, and moving towards specified objects or rooms. Our approach is based on generating skill-specific interventions and measuring changes in agent predictions. We present a detailed case study analyzing the behavior of a recent agent and then compare multiple agents in terms of skill-specific competency scores. This analysis suggests that biases from training have lasting effects on agent behavior and that existing models are able to ground simple referring expressions. Our comparisons between models show that skill-specific scores correlate with improvements in overall VLN task performance.

Vision-and-language navigation (VLN) is the task to enable an embodied agent to navigate to a remote location following the natural language instruction in real scenes. Most of the previous approaches utilize the entire features or object-centric features to represent navigable candidates. However, these representations are not efficient enough for an agent to perform actions to arrive the target location. As knowledge provides crucial information which is complementary to visible content, in this paper, we propose a Knowledge Enhanced Reasoning Model (KERM) to leverage knowledge to improve agent navigation ability. Specifically, we first retrieve facts (i.e., knowledge described by language descriptions) for the navigation views based on local regions from the constructed knowledge base. The retrieved facts range from properties of a single object (e.g., color, shape) to relationships between objects (e.g., action, spatial position), providing crucial information for VLN. We further present the KERM which contains the purification, fact-aware interaction, and instruction-guided aggregation modules to integrate visual, history, instruction, and fact features. The proposed KERM can automatically select and gather crucial and relevant cues, obtaining more accurate action prediction. Experimental results on the REVERIE, R2R, and SOON datasets demonstrate the effectiveness of the proposed method. The source code is available …

This paper deals with the problem of localizing objects in image and video datasets from visual exemplars. In particular, we focus on the challenging problem of egocentric visual query localization. We first identify grave implicit biases in current query-conditioned model design and visual query datasets. Then, we directly tackle such biases at both frame and object set levels. Concretely, our method solves these issues by expanding limited annotations and dynamically dropping object proposals during training. Additionally, we propose a novel transformer-based module that allows for object-proposal set context to be considered while incorporating query information. We name our module Conditioned Contextual Transformer or CocoFormer. Our experiments show that the proposed adaptations improve egocentric query detection, leading to a better visual query localization system in both 2D and 3D configurations. Thus, we are able to improve frame-level detection performance from 26.28% to 31.26% in AP, which correspondingly improves the VQ2D and VQ3D localization scores by significant margins. Our improved context-aware query object detector ranked first and second in the VQ2D and VQ3D tasks in the 2nd Ego4D challenge. In addition, we showcase the relevance of our proposed model in the Few-Shot Detection (FSD) task, where we also achieve SOTA results.

Large-scale pre-training has brought unimodal fields such as computer vision and natural language processing to a new era. Following this trend, the size of multimodal learning models constantly increases, leading to an urgent need to reduce the massive computational cost of fine-tuning these models for downstream tasks. In this paper, we propose an efficient and flexible multimodal fusion method, namely PMF, tailored for fusing unimodally pretrained transformers. Specifically, we first present a modular multimodal fusion framework that exhibits high flexibility and facilitates mutual interactions among different modalities. In addition, we disentangle vanilla prompts into three types in order to learn different optimizing objectives for multimodal learning. It is also worth noting that we propose to add prompt vectors only on the deep layers of the unimodal transformers, thus significantly reducing the training memory usage. Experiment results show that our proposed method achieves comparable performance to several other multimodal finetuning methods with less than 3% trainable parameters and up to 66% saving of training memory usage.

Grounding object properties and relations in 3D scenes is a prerequisite for a wide range of artificial intelligence tasks, such as visually grounded dialogues and embodied manipulation. However, the variability of the 3D domain induces two fundamental challenges: 1) the expense of labeling and 2) the complexity of 3D grounded language. Hence, essential desiderata for models are to be data-efficient, generalize to different data distributions and tasks with unseen semantic forms, as well as ground complex language semantics (e.g., view-point anchoring and multi-object reference). To address these challenges, we propose NS3D, a neuro-symbolic framework for 3D grounding. NS3D translates language into programs with hierarchical structures by leveraging large language-to-code models. Different functional modules in the programs are implemented as neural networks. Notably, NS3D extends prior neuro-symbolic visual reasoning methods by introducing functional modules that effectively reason about high-arity relations (i.e., relations among more than two objects), key in disambiguating objects in complex 3D scenes. Modular and compositional architecture enables NS3D to achieve state-of-the-art results on the ReferIt3D view-dependence task, a 3D referring expression comprehension benchmark. Importantly, NS3D shows significantly improved performance on settings of data-efficiency and generalization, and demonstrate zero-shot transfer to an unseen 3D question-answering task.

Transformers have been recently utilized for vision and language tasks successfully. For example, recent image and language models with more than 200M parameters have been proposed to learn visual grounding in the pre-training step and show impressive results on downstream vision and language tasks. On the other hand, there exists a large amount of computational redundancy in these large models which skips their run-time efficiency. To address this problem, we propose dynamic inference for grounding based vision and language models conditioned on the input image-text pair. We first design an approach to dynamically skip multihead self-attention and feed forward network layers across two backbones and multimodal network. Additionally, we propose dynamic token pruning and fusion for two backbones. In particular, we remove redundant tokens at different levels of the backbones and fuse the image tokens with the language tokens in an adaptive manner. To learn policies for dynamic inference, we train agents using reinforcement learning. In this direction, we replace the CNN backbone in a recent grounding-based vision and language model, MDETR, with a vision transformer and call it ViTMDETR. Then, we apply our dynamic inference method to ViTMDETR, called D-ViTDMETR, and perform experiments on image-language tasks. Our results show …

This paper focuses on analyzing and improving the commonsense ability of recent popular vision-language (VL) models. Despite the great success, we observe that existing VL-models still lack commonsense knowledge/reasoning ability (e.g., “Lemons are sour”), which is a vital component towards artificial general intelligence. Through our analysis, we find one important reason is that existing large-scale VL datasets do not contain much commonsense knowledge, which motivates us to improve the commonsense of VL-models from the data perspective. Rather than collecting a new VL training dataset, we propose a more scalable strategy, i.e., “Data Augmentation with kNowledge graph linearization for CommonsensE capability” (DANCE). It can be viewed as one type of data augmentation technique, which can inject commonsense knowledge into existing VL datasets on the fly during training. More specifically, we leverage the commonsense knowledge graph (e.g., ConceptNet) and create variants of text description in VL datasets via bidirectional sub-graph sequentialization. For better commonsense evaluation, we further propose the first retrieval-based commonsense diagnostic benchmark. By conducting extensive experiments on some representative VL-models, we demonstrate that our DANCE technique is able to significantly improve the commonsense ability while maintaining the performance on vanilla retrieval tasks.

VQA Natural Language Explanation (VQA-NLE) task aims to explain the decision-making process of VQA models in natural language. Unlike traditional attention or gradient analysis, free-text rationales can be easier to understand and gain users’ trust. Existing methods mostly use post-hoc or self-rationalization models to obtain a plausible explanation. However, these frameworks are bottlenecked by the following challenges: 1) the reasoning process cannot be faithfully responded to and suffer from the problem of logical inconsistency. 2) Human-annotated explanations are expensive and time-consuming to collect. In this paper, we propose a new Semi-Supervised VQA-NLE via Self-Critical Learning (S3C), which evaluates the candidate explanations by answering rewards to improve the logical consistency between answers and rationales. With a semi-supervised learning framework, the S3C can benefit from a tremendous amount of samples without human-annotated explanations. A large number of automatic measures and human evaluations all show the effectiveness of our method. Meanwhile, the framework achieves a new state-of-the-art performance on the two VQA-NLE datasets.

Vision and Language (VL) models have demonstrated remarkable zero-shot performance in a variety of tasks. However, some aspects of complex language understanding still remain a challenge. We introduce the collective notion of Structured Vision & Language Concepts (SVLC) which includes object attributes, relations, and states which are present in the text and visible in the image. Recent studies have shown that even the best VL models struggle with SVLC. A possible way of fixing this issue is by collecting dedicated datasets for teaching each SVLC type, yet this might be expensive and time-consuming. Instead, we propose a more elegant data-driven approach for enhancing VL models’ understanding of SVLCs that makes more effective use of existing VL pre-training datasets and does not require any additional data. While automatic understanding of image structure still remains largely unsolved, language structure is much better modeled and understood, allowing for its effective utilization in teaching VL models. In this paper, we propose various techniques based on language structure understanding that can be used to manipulate the textual part of off-the-shelf paired VL datasets. VL models trained with the updated data exhibit a significant improvement of up to 15% in their SVLC understanding with only a …

In the fashion domain, there exists a variety of vision-and-language (V+L) tasks, including cross-modal retrieval, text-guided image retrieval, multi-modal classification, and image captioning. They differ drastically in each individual input/output format and dataset size. It has been common to design a task-specific model and fine-tune it independently from a pre-trained V+L model (e.g., CLIP). This results in parameter inefficiency and inability to exploit inter-task relatedness. To address such issues, we propose a novel FAshion-focused Multi-task Efficient learning method for Vision-and-Language tasks (FAME-ViL) in this work. Compared with existing approaches, FAME-ViL applies a single model for multiple heterogeneous fashion tasks, therefore being much more parameter-efficient. It is enabled by two novel components: (1) a task-versatile architecture with cross-attention adapters and task-specific adapters integrated into a unified V+L model, and (2) a stable and effective multi-task training strategy that supports learning from heterogeneous data and prevents negative transfer. Extensive experiments on four fashion tasks show that our FAME-ViL can save 61.5% of parameters over alternatives, while significantly outperforming the conventional independently trained single-task models. Code is available at https://github.com/BrandonHanx/FAME-ViL.

Referring Expression Comprehension (REC) is a task of grounding the referent based on an expression, and its development is greatly limited by expensive instance-level annotations. Most existing weakly supervised methods are built based on two-stage detection networks, which are computationally expensive. In this paper, we resort to the efficient one-stage detector and propose a novel weakly supervised model called RefCLIP. Specifically, RefCLIP redefines weakly supervised REC as an anchor-text matching problem, which can avoid the complex post-processing in existing methods. To achieve weakly supervised learning, we introduce anchor-based contrastive loss to optimize RefCLIP via numerous anchor-text pairs. Based on RefCLIP, we further propose the first model-agnostic weakly supervised training scheme for existing REC models, where RefCLIP acts as a mature teacher to generate pseudo-labels for teaching common REC models. With our careful designs, this scheme can even help existing REC models achieve better weakly supervised performance than RefCLIP, e.g., TransVG and SimREC. To validate our approaches, we conduct extensive experiments on four REC benchmarks, i.e., RefCOCO, RefCOCO+, RefCOCOg and ReferItGame. Experimental results not only report our significant performance gains over existing weakly supervised models, e.g., +24.87% on RefCOCO, but also show the 5x faster inference speed. Project: https://refclip.github.io.

Despite the remarkable success of foundation models, their task-specific fine-tuning paradigm makes them inconsistent with the goal of general perception modeling. The key to eliminating this inconsistency is to use generalist models for general task modeling. However, existing attempts at generalist models are inadequate in both versatility and performance. In this paper, we propose Uni-Perceiver v2, which is the first generalist model capable of handling major large-scale vision and vision-language tasks with competitive performance. Specifically, images are encoded as general region proposals, while texts are encoded via a Transformer-based language model. The encoded representations are transformed by a task-agnostic decoder. Different tasks are formulated as a unified maximum likelihood estimation problem. We further propose an effective optimization technique named Task-Balanced Gradient Normalization to ensure stable multi-task learning with an unmixed sampling strategy, which is helpful for tasks requiring large batch-size training. After being jointly trained on various tasks, Uni-Perceiver v2 is capable of directly handling downstream tasks without any task-specific adaptation. Results show that Uni-Perceiver v2 outperforms all existing generalist models in both versatility and performance. Meanwhile, compared with the commonly-recognized strong baselines that require tasks-specific fine-tuning, Uni-Perceiver v2 achieves competitive performance on a broad range of vision and vision-language …

Everyone is unique. Given the same visual stimuli, people’s attention is driven by both salient visual cues and their own inherent preferences. Knowledge of visual preferences not only facilitates understanding of fine-grained attention patterns of diverse users, but also has the potential of benefiting the development of customized applications. Nevertheless, existing saliency models typically limit their scope to attention as it applies to the general population and ignore the variability between users’ behaviors. In this paper, we identify the critical roles of visual preferences in attention modeling, and for the first time study the problem of user-aware saliency modeling. Our work aims to advance attention research from three distinct perspectives: (1) We present a new model with the flexibility to capture attention patterns of various combinations of users, so that we can adaptively predict personalized attention, user group attention, and general saliency at the same time with one single model; (2) To augment models with knowledge about the composition of attention from different users, we further propose a principled learning method to understand visual attention in a progressive manner; and (3) We carry out extensive analyses on publicly available saliency datasets to shed light on the roles of visual preferences. …

Attribution methods are a popular class of explainability methods that use heatmaps to depict the most important areas of an image that drive a model decision. Nevertheless, recent work has shown that these methods have limited utility in practice, presumably because they only highlight the most salient parts of an image (i.e., “where” the model looked) and do not communicate any information about “what” the model saw at those locations. In this work, we try to fill in this gap with Craft -- a novel approach to identify both “what” and “where” by generating concept-based explanations. We introduce 3 new ingredients to the automatic concept extraction literature: (i) a recursive strategy to detect and decompose concepts across layers, (ii) a novel method for a more faithful estimation of concept importance using Sobol indices, and (iii) the use of implicit differentiation to unlock Concept Attribution Maps. We conduct both human and computer vision experiments to demonstrate the benefits of the proposed approach. We show that our recursive decomposition generates meaningful and accurate concepts and that the proposed concept importance estimation technique is more faithful to the model than previous methods. When evaluating the usefulness of the method for human experimenters on …

Many visual recognition models are evaluated only on their classification accuracy, a metric for which they obtain strong performance. In this paper, we investigate whether computer vision models can also provide correct rationales for their predictions. We propose a “doubly right” object recognition benchmark, where the metric requires the model to simultaneously produce both the right labels as well as the right rationales. We find that state-of-the-art visual models, such as CLIP, often provide incorrect rationales for their categorical predictions. However, by transferring the rationales from language models into visual representations through a tailored dataset, we show that we can learn a “why prompt,” which adapts large visual representations to produce correct rationales. Visualizations and empirical experiments show that our prompts significantly improve performance on doubly right object recognition, in addition to zero-shot transfer to unseen tasks and datasets.

Human sketch has already proved its worth in various visual understanding tasks (e.g., retrieval, segmentation, image-captioning, etc). In this paper, we reveal a new trait of sketches -- that they are also salient. This is intuitive as sketching is a natural attentive process at its core. More specifically, we aim to study how sketches can be used as a weak label to detect salient objects present in an image. To this end, we propose a novel method that emphasises on how “salient object” could be explained by hand-drawn sketches. To accomplish this, we introduce a photo-to-sketch generation model that aims to generate sequential sketch coordinates corresponding to a given visual photo through a 2D attention mechanism. Attention maps accumulated across the time steps give rise to salient regions in the process. Extensive quantitative and qualitative experiments prove our hypothesis and delineate how our sketch-based saliency detection model gives a competitive performance compared to the state-of-the-art.

Interpretable methods based on prototypical patches recognize various components in an image in order to explain their reasoning to humans. However, existing prototype-based methods can learn prototypes that are not in line with human visual perception, i.e., the same prototype can refer to different concepts in the real world, making interpretation not intuitive. Driven by the principle of explainability-by-design, we introduce PIP-Net (Patch-based Intuitive Prototypes Network): an interpretable image classification model that learns prototypical parts in a self-supervised fashion which correlate better with human vision. PIP-Net can be interpreted as a sparse scoring sheet where the presence of a prototypical part in an image adds evidence for a class. The model can also abstain from a decision for out-of-distribution data by saying “I haven’t seen this before”. We only use image-level labels and do not rely on any part annotations. PIP-Net is globally interpretable since the set of learned prototypes shows the entire reasoning of the model. A smaller local explanation locates the relevant prototypes in one image. We show that our prototypes correlate with ground-truth object parts, indicating that PIP-Net closes the “semantic gap” between latent space and pixel space. Hence, our PIP-Net with interpretable prototypes enables users to …

The sketch community has faced up to its unique challenges over the years, that of data scarcity however still remains the most significant to date. This lack of sketch data has imposed on the community a few “peculiar” design choices -- the most representative of them all is perhaps the coerced utilisation of photo-based pre-training (i.e., no sketch), for many core tasks that otherwise dictates specific sketch understanding. In this paper, we ask just the one question -- can we make such photo-based pre-training, to actually benefit sketch? Our answer lies in cultivating the topology of photo data learned at pre-training, and use that as a “free” source of supervision for downstream sketch tasks. In particular, we use fine-grained sketch-based image retrieval (FG-SBIR), one of the most studied and data-hungry sketch tasks, to showcase our new perspective on pre-training. In this context, the topology-informed supervision learned from photos act as a constraint that take effect at every fine-tuning step -- neighbouring photos in the pre-trained model remain neighbours under each FG-SBIR updates. We further portray this neighbourhood consistency constraint as a photo ranking problem and formulate it into a neat cross-modal triplet loss. We also show how this target is …

In this paper, we leverage CLIP for zero-shot sketch based image retrieval (ZS-SBIR). We are largely inspired by recent advances on foundation models and the unparalleled generalisation ability they seem to offer, but for the first time tailor it to benefit the sketch community. We put forward novel designs on how best to achieve this synergy, for both the category setting and the fine-grained setting (“all”). At the very core of our solution is a prompt learning setup. First we show just via factoring in sketch-specific prompts, we already have a category-level ZS-SBIR system that overshoots all prior arts, by a large margin (24.8%) - a great testimony on studying the CLIP and ZS-SBIR synergy. Moving onto the fine-grained setup is however trickier, and requires a deeper dive into this synergy. For that, we come up with two specific designs to tackle the fine-grained matching nature of the problem: (i) an additional regularisation loss to ensure the relative separation between sketches and photos is uniform across categories, which is not the case for the gold standard standalone triplet loss, and (ii) a clever patch shuffling technique to help establishing instance-level structural correspondences between sketch-photo pairs. With these designs, we again …

This paper presents a method that effectively combines two prevalent visual recognition methods, i.e., image classification and contrastive language-image pre-training, dubbed iCLIP. Instead of naive multi-task learning that use two separate heads for each task, we fuse the two tasks in a deep fashion that adapts the image classification to share the same formula and the same model weights with the language-image pre-training. To further bridge these two tasks, we propose to enhance the category names in image classification tasks using external knowledge, such as their descriptions in dictionaries. Extensive experiments show that the proposed method combines the advantages of two tasks well: the strong discrimination ability in image classification tasks due to the clear and clean category labels, and the good zero-shot ability in CLIP tasks ascribed to the richer semantics in the text descriptions. In particular, it reaches 82.9% top-1 accuracy on IN-1K, and surpasses CLIPby 1.8%, with similar model size, on zero-shot recognition of Kornblith 12-dataset benchmark. The code and models are publicly available at https://github.com/weiyx16/iCLIP.

Text-to-image person retrieval aims to identify the target person based on a given textual description query. The primary challenge is to learn the mapping of visual and textual modalities into a common latent space. Prior works have attempted to address this challenge by leveraging separately pre-trained unimodal models to extract visual and textual features. However, these approaches lack the necessary underlying alignment capabilities required to match multimodal data effectively. Besides, these works use prior information to explore explicit part alignments, which may lead to the distortion of intra-modality information. To alleviate these issues, we present IRRA: a cross-modal Implicit Relation Reasoning and Aligning framework that learns relations between local visual-textual tokens and enhances global image-text matching without requiring additional prior supervision. Specifically, we first design an Implicit Relation Reasoning module in a masked language modeling paradigm. This achieves cross-modal interaction by integrating the visual cues into the textual tokens with a cross-modal multimodal interaction encoder. Secondly, to globally align the visual and textual embeddings, Similarity Distribution Matching is proposed to minimize the KL divergence between image-text similarity distributions and the normalized label matching distributions. The proposed method achieves new state-of-the-art results on all three public datasets, with a notable margin …

We propose a visual-linguistic representation learning approach within a self-supervised learning framework by introducing a new operation, loss, and data augmentation strategy. First, we generate diverse features for the image-text matching (ITM) task via soft-masking the regions in an image, which are most relevant to a certain word in the corresponding caption, instead of completely removing them. Since our framework relies only on image-caption pairs with no fine-grained annotations, we identify the relevant regions to each word by computing the word-conditional visual attention using multi-modal encoder. Second, we encourage the model to focus more on hard but diverse examples by proposing a focal loss for the image-text contrastive learning (ITC) objective, which alleviates the inherent limitations of overfitting and bias issues. Last, we perform multi-modal data augmentations for self-supervised learning via mining various examples by masking texts and rendering distortions on images. We show that the combination of these three innovations is effective for learning a pretrained model, leading to outstanding performance on multiple vision-language downstream tasks.

Prompt tuning has been employed as an efficient way to adapt large vision-language pre-trained models (e.g. CLIP) to various downstream tasks in data-limited or label-limited settings. Nonetheless, visual data (e.g., images) is by default prerequisite for learning prompts in existing methods. In this work, we advocate that the effectiveness of image-text contrastive learning in aligning the two modalities (for training CLIP) further makes it feasible to treat texts as images for prompt tuning and introduce TaI prompting. In contrast to the visual data, text descriptions are easy to collect, and their class labels can be directly derived. Particularly, we apply TaI prompting to multi-label image recognition, where sentences in the wild serve as alternatives to images for prompt tuning. Moreover, with TaI, double-grained prompt tuning (TaI-DPT) is further presented to extract both coarse-grained and fine-grained embeddings for enhancing the multi-label recognition performance. Experimental results show that our proposed TaI-DPT outperforms zero-shot CLIP by a large margin on multiple benchmarks, e.g., MS-COCO, VOC2007, and NUS-WIDE, while it can be combined with existing methods of prompting from images to improve recognition performance further. The code is released at https://github.com/guozix/TaI-DPT.

Scaling up neural networks has led to remarkable performance across a wide range of tasks. Moreover, performance often follows reliable scaling laws as a function of training set size, model size, and compute, which offers valuable guidance as large-scale experiments are becoming increasingly expensive. However, previous work on scaling laws has primarily used private data & models or focused on uni-modal language or vision learning. To address these limitations, we investigate scaling laws for contrastive language-image pre-training (CLIP) with the public LAION dataset and the open-source OpenCLIP repository. Our large-scale experiments involve models trained on up to two billion image-text pairs and identify power law scaling for multiple downstream tasks including zero-shot classification, retrieval, linear probing, and end-to-end fine-tuning. We find that the training distribution plays a key role in scaling laws as the OpenAI and OpenCLIP models exhibit different scaling behavior despite identical model architectures and similar training recipes. We open-source our evaluation workflow and all models, including the largest public CLIP models, to ensure reproducibility and make scaling laws research more accessible. Source code and instructions to reproduce this study is available at https://github.com/LAION-AI/scaling-laws-openclip.

Image-text retrieval is a fundamental task to bridge vision and language by exploiting various strategies to fine-grained alignment between regions and words. This is still tough mainly because of one-to-many correspondence, where a set of matches from another modality can be accessed by a random query. While existing solutions to this problem including multi-point mapping, probabilistic distribution, and geometric embedding have made promising progress, one-to-many correspondence is still under-explored. In this work, we develop a Multilateral Semantic Relations Modeling (termed MSRM) for image-text retrieval to capture the one-to-many correspondence between multiple samples and a given query via hypergraph modeling. Specifically, a given query is first mapped as a probabilistic embedding to learn its true semantic distribution based on Mahalanobis distance. Then each candidate instance in a mini-batch is regarded as a hypergraph node with its mean semantics while a Gaussian query is modeled as a hyperedge to capture the semantic correlations beyond the pair between candidate points and the query. Comprehensive experimental results on two widely used datasets demonstrate that our MSRM method can outperform state-of-the-art methods in the settlement of multiple matches while still maintaining the comparable performance of instance-level matching. Our codes and checkpoints will be released soon.

Recent advances in image captioning have focused on scaling the data and model size, substantially increasing the cost of pre-training and finetuning. As an alternative to large models, we present SmallCap, which generates a caption conditioned on an input image and related captions retrieved from a datastore. Our model is lightweight and fast to train as the only learned parameters are in newly introduced cross-attention layers between a pre-trained CLIP encoder and GPT-2 decoder. SmallCap can transfer to new domains without additional finetuning and can exploit large-scale data in a training-free fashion since the contents of the datastore can be readily replaced. Our experiments show that SmallCap, trained only on COCO, has competitive performance on this benchmark, and also transfers to other domains without retraining, solely through retrieval from target-domain data. Further improvement is achieved through the training-free exploitation of diverse human-labeled and web data, which proves effective for a range of domains, including the nocaps benchmark, designed to test generalization to unseen visual concepts.

Pre-training of deep convolutional neural networks (DCNNs) plays a crucial role in the field of visual sentiment analysis (VSA). Most proposed methods employ the off-the-shelf backbones pre-trained on large-scale object classification datasets (i.e., ImageNet). While it boosts performance for a big margin against initializing model states from random, we argue that DCNNs simply pre-trained on ImageNet may excessively focus on recognizing objects, but failed to provide high-level concepts in terms of sentiment. To address this long-term overlooked problem, we propose a sentiment-oriented pre-training method that is built upon human visual sentiment perception (VSP) mechanism. Specifically, we factorize the process of VSP into three steps, namely stimuli taking, holistic organizing, and high-level perceiving. From imitating each VSP step, a total of three models are separately pre-trained via our devised sentiment-aware tasks that contribute to excavating sentiment-discriminated representations. Moreover, along with our elaborated multi-model amalgamation strategy, the prior knowledge learned from each perception step can be effectively transferred into a single target model, yielding substantial performance gains. Finally, we verify the superiorities of our proposed method over extensive experiments, covering mainstream VSA tasks from single-label learning (SLL), multi-label learning (MLL), to label distribution learning (LDL). Experiment results demonstrate that our proposed method …

Pretraining visual models on web-scale image-caption datasets has recently emerged as a powerful alternative to traditional pretraining on image classification data. Image-caption datasets are more “open-domain”, containing broader scene types and vocabulary words, and result in models that have strong performance in few- and zero-shot recognition tasks. However large-scale classification datasets can provide fine-grained categories with a balanced label distribution. In this work, we study a pretraining strategy that uses both classification and caption datasets to unite their complementary benefits. First, we show that naively unifying the datasets results in sub-optimal performance in downstream zero-shot recognition tasks, as the model is affected by dataset bias: the coverage of image domains and vocabulary words is different in each dataset. We address this problem with novel Prefix Conditioning, a simple yet effective method that helps disentangle dataset biases from visual concepts. This is done by introducing prefix tokens that inform the language encoder of the input data type (e.g., classification vs caption) at training time. Our approach allows the language encoder to learn from both datasets while also tailoring feature extraction to each dataset. Prefix conditioning is generic and can be easily integrated into existing VL pretraining objectives, such as CLIP or …

Existing image captioning methods are under the assumption that the training and testing data are from the same domain or that the data from the target domain (i.e., the domain that testing data lie in) are accessible. However, this assumption is invalid in real-world applications where the data from the target domain is inaccessible. In this paper, we introduce a new setting called Domain Generalization for Image Captioning (DGIC), where the data from the target domain is unseen in the learning process. We first construct a benchmark dataset for DGIC, which helps us to investigate models’ domain generalization (DG) ability on unseen domains. With the support of the new benchmark, we further propose a new framework called language-guided semantic metric learning (LSML) for the DGIC setting. Experiments on multiple datasets demonstrate the challenge of the task and the effectiveness of our newly proposed benchmark and LSML framework.

This work investigates dataset vectorization for two dataset-level tasks: assessing training set suitability and test set difficulty. The former measures how suitable a training set is for a target domain, while the latter studies how challenging a test set is for a learned model. Central of the two tasks is measuring the underlying relationship between datasets. This needs a desirable dataset vectorization scheme, which should preserve as much discriminative dataset information as possible so that the distance between the resulting dataset vectors can reflect dataset-to-dataset similarity. To this end, we propose a bag-of-prototypes (BoP) dataset representation that extends the image level bag consisting of patch descriptors to dataset-level bag consisting of semantic prototypes. Specifically, we develop a codebook consisting of K prototypes clustered from a reference dataset. Given a dataset to be encoded, we quantize each of its image features to a certain prototype in the codebook and obtain a K-dimensional histogram feature. Without assuming access to dataset labels, the BoP representation provides rich characterization of dataset semantic distribution. Further, BoP representations cooperates well with Jensen-Shannon divergence for measuring dataset-to-dataset similarity. Albeit very simple, BoP consistently shows its advantage over existing representations on a series of benchmarks for two dataset-level …

Supervised crowd counting relies heavily on costly manual labeling, which is difficult and expensive, especially in dense scenes. To alleviate the problem, we propose a novel unsupervised framework for crowd counting, named CrowdCLIP. The core idea is built on two observations: 1) the recent contrastive pre-trained vision-language model (CLIP) has presented impressive performance on various downstream tasks; 2) there is a natural mapping between crowd patches and count text. To the best of our knowledge, CrowdCLIP is the first to investigate the vision-language knowledge to solve the counting problem. Specifically, in the training stage, we exploit the multi-modal ranking loss by constructing ranking text prompts to match the size-sorted crowd patches to guide the image encoder learning. In the testing stage, to deal with the diversity of image patches, we propose a simple yet effective progressive filtering strategy to first select the highly potential crowd patches and then map them into the language space with various counting intervals. Extensive experiments on five challenging datasets demonstrate that the proposed CrowdCLIP achieves superior performance compared to previous unsupervised state-of-the-art counting methods. Notably, CrowdCLIP even surpasses some popular fully-supervised methods under the cross-dataset setting. The source code will be available at https://github.com/dk-liang/CrowdCLIP.

Establishing pixel-level matches between image pairs is vital for a variety of computer vision applications. However, achieving robust image matching remains challenging because CNN extracted descriptors usually lack discriminative ability in texture-less regions and keypoint detectors are only good at identifying keypoints with a specific level of structure. To deal with these issues, a novel image matching method is proposed by Jointly Learning Hierarchical Detectors and Contextual Descriptors via Agent-based Transformers (D2Former), including a contextual feature descriptor learning (CFDL) module and a hierarchical keypoint detector learning (HKDL) module. The proposed D2Former enjoys several merits. First, the proposed CFDL module can model long-range contexts efficiently and effectively with the aid of designed descriptor agents. Second, the HKDL module can generate keypoint detectors in a hierarchical way, which is helpful for detecting keypoints with diverse levels of structures. Extensive experimental results on four challenging benchmarks show that our proposed method significantly outperforms state-of-the-art image matching methods.

Scene graph generation (SGG) aims to abstract an image into a graph structure, by representing objects as graph nodes and their relations as labeled edges. However, two knotty obstacles limit the practicability of current SGG methods in real-world scenarios: 1) training SGG models requires time-consuming ground-truth annotations, and 2) the closed-set object categories make the SGG models limited in their ability to recognize novel objects outside of training corpora. To address these issues, we novelly exploit a powerful pre-trained visual-semantic space (VSS) to trigger language-supervised and open-vocabulary SGG in a simple yet effective manner. Specifically, cheap scene graph supervision data can be easily obtained by parsing image language descriptions into semantic graphs. Next, the noun phrases on such semantic graphs are directly grounded over image regions through region-word alignment in the pre-trained VSS. In this way, we enable open-vocabulary object detection by performing object category name grounding with a text prompt in this VSS. On the basis of visually-grounded objects, the relation representations are naturally built for relation recognition, pursuing open-vocabulary SGG. We validate our proposed approach with extensive experiments on the Visual Genome benchmark across various SGG scenarios (i.e., supervised / language-supervised, closed-set / open-vocabulary). Consistent superior performances are …

Recent state-of-the-art methods for HOI detection typically build on transformer architectures with two decoder branches, one for human-object pair detection and the other for interaction classification. Such disentangled transformers, however, may suffer from insufficient context exchange between the branches and lead to a lack of context information for relational reasoning, which is critical in discovering HOI instances. In this work, we propose the multiplex relation network (MUREN) that performs rich context exchange between three decoder branches using unary, pairwise, and ternary relations of human, object, and interaction tokens. The proposed method learns comprehensive relational contexts for discovering HOI instances, achieving state-of-the-art performance on two standard benchmarks for HOI detection, HICO-DET and V-COCO.

In this paper, we consider the problem of open-vocabulary semantic segmentation (OVS), which aims to segment objects of arbitrary classes instead of pre-defined, closed-set categories. The main contributions are as follows: First, we propose a transformer-based model for OVS, termed as OVSegmentor, which only exploits web-crawled image-text pairs for pre-training without using any mask annotations. OVSegmentor assembles the image pixels into a set of learnable group tokens via a slot-attention based binding module, and aligns the group tokens to the corresponding caption embedding. Second, we propose two proxy tasks for training, namely masked entity completion and cross-image mask consistency. The former aims to infer all masked entities in the caption given the group tokens, that enables the model to learn fine-grained alignment between visual groups and text entities. The latter enforces consistent mask predictions between images that contain shared entities, which encourages the model to learn visual invariance. Third, we construct CC4M dataset for pre-training by filtering CC12M with frequently appeared entities, which significantly improves training efficiency. Fourth, we perform zero-shot transfer on three benchmark datasets, PASCAL VOC 2012, PASCAL Context, and COCO Object. Our model achieves superior segmentation results over the state-of-the-art method by using only 3% data (4M …

This paper presents a new framework for open-vocabulary semantic segmentation with the pre-trained vision-language model, named SAN. Our approach models the semantic segmentation task as a region recognition problem. A side network is attached to a frozen CLIP model with two branches: one for predicting mask proposals, and the other for predicting attention bias which is applied in the CLIP model to recognize the class of masks. This decoupled design has the benefit CLIP in recognizing the class of mask proposals. Since the attached side network can reuse CLIP features, it can be very light. In addition, the entire network can be trained end-to-end, allowing the side network to be adapted to the frozen CLIP model, which makes the predicted mask proposals CLIP-aware. Our approach is fast, accurate, and only adds a few additional trainable parameters. We evaluate our approach on multiple semantic segmentation benchmarks. Our method significantly outperforms other counterparts, with up to 18 times fewer trainable parameters and 19 times faster inference speed. We hope our approach will serve as a solid baseline and help ease future research in open-vocabulary semantic segmentation.

We present ODISE: Open-vocabulary DIffusion-based panoptic SEgmentation, which unifies pre-trained text-image diffusion and discriminative models to perform open-vocabulary panoptic segmentation. Text-to-image diffusion models have the remarkable ability to generate high-quality images with diverse open-vocabulary language descriptions. This demonstrates that their internal representation space is highly correlated with open concepts in the real world. Text-image discriminative models like CLIP, on the other hand, are good at classifying images into open-vocabulary labels. We leverage the frozen internal representations of both these models to perform panoptic segmentation of any category in the wild. Our approach outperforms the previous state of the art by significant margins on both open-vocabulary panoptic and semantic segmentation tasks. In particular, with COCO training only, our method achieves 23.4 PQ and 30.0 mIoU on the ADE20K dataset, with 8.3 PQ and 7.9 mIoU absolute improvement over the previous state of the art. We open-source our code and models at https://github.com/NVlabs/ODISE.

Vision-language (VL) pre-training has recently gained much attention for its transferability and flexibility in novel concepts (e.g., cross-modality transfer) across various visual tasks. However, VL-driven segmentation has been under-explored, and the existing approaches still have the burden of acquiring additional training images or even segmentation annotations to adapt a VL model to downstream segmentation tasks. In this paper, we introduce a novel image-free segmentation task where the goal is to perform semantic segmentation given only a set of the target semantic categories, but without any task-specific images and annotations. To tackle this challenging task, our proposed method, coined IFSeg, generates VL-driven artificial image-segmentation pairs and updates a pre-trained VL model to a segmentation task. We construct this artificial training data by creating a 2D map of random semantic categories and another map of their corresponding word tokens. Given that a pre-trained VL model projects visual and text tokens into a common space where tokens that share the semantics are located closely, this artificially generated word map can replace the real image inputs for such a VL model. Through an extensive set of experiments, our model not only establishes an effective baseline for this novel task but also demonstrates strong performances …

Learning a generalizable object manipulation policy is vital for an embodied agent to work in complex real-world scenes. Parts, as the shared components in different object categories, have the potential to increase the generalization ability of the manipulation policy and achieve cross-category object manipulation. In this work, we build the first large-scale, part-based cross-category object manipulation benchmark, PartManip, which is composed of 11 object categories, 494 objects, and 1432 tasks in 6 task classes. Compared to previous work, our benchmark is also more diverse and realistic, i.e., having more objects and using sparse-view point cloud as input without oracle information like part segmentation. To tackle the difficulties of vision-based policy learning, we first train a state-based expert with our proposed part-based canonicalization and part-aware rewards, and then distill the knowledge to a vision-based student. We also find an expressive backbone is essential to overcome the large diversity of different objects. For cross-category generalization, we introduce domain adversarial learning for domain-invariant feature extraction. Extensive experiments in simulation show that our learned policy can outperform other methods by a large margin, especially on unseen object categories. We also demonstrate our method can successfully manipulate novel objects in the real world.

Universal Image Segmentation is not a new concept.Past attempts to unify image segmentation include scene parsing, panoptic segmentation, and, more recently, new panoptic architectures. However, such panoptic architectures do not truly unify image segmentation because they need to be trained individually on the semantic, instance, or panoptic segmentation to achieve the best performance. Ideally, a truly universal framework should be trained only once and achieve SOTA performance across all three image segmentation tasks. To that end, we propose OneFormer, a universal image segmentation framework that unifies segmentation with a multi-task train-once design. We first propose a task-conditioned joint training strategy that enables training on ground truths of each domain (semantic, instance, and panoptic segmentation) within a single multi-task training process. Secondly, we introduce a task token to condition our model on the task at hand, making our model task-dynamic to support multi-task training and inference. Thirdly, we propose using a query-text contrastive loss during training to establish better inter-task and inter-class distinctions. Notably, our single OneFormer model outperforms specialized Mask2Former models across all three segmentation tasks on ADE20k, Cityscapes, and COCO, despite the latter being trained on each task individually. We believe OneFormer is a significant step towards making image …

Thanks to the impressive progress of large-scale vision-language pretraining, recent recognition models can classify arbitrary objects in a zero-shot and open-set manner, with a surprisingly high accuracy. However, translating this success to semantic segmentation is not trivial, because this dense prediction task requires not only accurate semantic understanding but also fine shape delineation and existing vision-language models are trained with image-level language descriptions. To bridge this gap, we pursue shape-aware zero-shot semantic segmentation in this study. Inspired by classical spectral methods in the image segmentation literature, we propose to leverage the eigen vectors of Laplacian matrices constructed with self-supervised pixel-wise features to promote shape-awareness. Despite that this simple and effective technique does not make use of the masks of seen classes at all, we demonstrate that it out-performs a state-of-the-art shape-aware formulation that aligns ground truth and predicted edges during training. We also delve into the performance gains achieved on different datasets using different backbones and draw several interesting and conclusive observations: the benefits of promoting shape-awareness highly relates to mask compactness and language embedding locality. Finally, our method sets new state-of-the-art performance for zero-shot semantic segmentation on both Pascal and COCO, with significant margins. Code and models will be …

Continual learning for segmentation has recently seen increasing interest. However, all previous works focus on narrow semantic segmentation and disregard panoptic segmentation, an important task with real-world impacts. In this paper, we present the first continual learning model capable of operating on both semantic and panoptic segmentation. Inspired by recent transformer approaches that consider segmentation as a mask-classification problem, we design CoMFormer. Our method carefully exploits the properties of transformer architectures to learn new classes over time. Specifically, we propose a novel adaptive distillation loss along with a mask-based pseudo-labeling technique to effectively prevent forgetting. To evaluate our approach, we introduce a novel continual panoptic segmentation benchmark on the challenging ADE20K dataset. Our CoMFormer outperforms all the existing baselines by forgetting less old classes but also learning more effectively new classes. In addition, we also report an extensive evaluation in the large-scale continual semantic segmentation scenario showing that CoMFormer also significantly outperforms state-of-the-art methods.

Referring Expression Segmentation (RES) can facilitate pixel-level semantic alignment between vision and language. Most of the existing RES approaches require massive pixel-level annotations, which are expensive and exhaustive. In this paper, we propose a new partially supervised training paradigm for RES, i.e., training using abundant referring bounding boxes and only a few (e.g., 1%) pixel-level referring masks. To maximize the transferability from the REC model, we construct our model based on the point-based sequence prediction model. We propose the co-content teacher-forcing to make the model explicitly associate the point coordinates (scale values) with the referred spatial features, which alleviates the exposure bias caused by the limited segmentation masks. To make the most of referring bounding box annotations, we further propose the resampling pseudo points strategy to select more accurate pseudo-points as supervision. Extensive experiments show that our model achieves 52.06% in terms of accuracy (versus 58.93% in fully supervised setting) on RefCOCO+@testA, when only using 1% of the mask annotations.

An increasing amount of applications rely on data-driven models that are deployed for perception tasks across a sequence of scenes. Due to the mismatch between training and deployment data, adapting the model on the new scenes is often crucial to obtain good performance. In this work, we study continual multi-scene adaptation for the task of semantic segmentation, assuming that no ground-truth labels are available during deployment and that performance on the previous scenes should be maintained. We propose training a Semantic-NeRF network for each scene by fusing the predictions of a segmentation model and then using the view-consistent rendered semantic labels as pseudo-labels to adapt the model. Through joint training with the segmentation model, the Semantic-NeRF model effectively enables 2D-3D knowledge transfer. Furthermore, due to its compact size, it can be stored in a long-term memory and subsequently used to render data from arbitrary viewpoints to reduce forgetting. We evaluate our approach on ScanNet, where we outperform both a voxel-based baseline and a state-of-the-art unsupervised domain adaptation method.

In this paper we present Mask DINO, a unified object detection and segmentation framework. Mask DINO extends DINO (DETR with Improved Denoising Anchor Boxes) by adding a mask prediction branch which supports all image segmentation tasks (instance, panoptic, and semantic). It makes use of the query embeddings from DINO to dot-product a high-resolution pixel embedding map to predict a set of binary masks. Some key components in DINO are extended for segmentation through a shared architecture and training process. Mask DINO is simple, efficient, scalable, and benefits from joint large-scale detection and segmentation datasets. Our experiments show that Mask DINO significantly outperforms all existing specialized segmentation methods, both on a ResNet-50 backbone and a pre-trained model with SwinL backbone. Notably, Mask DINO establishes the best results to date on instance segmentation (54.5 AP on COCO), panoptic segmentation (59.4 PQ on COCO), and semantic segmentation (60.8 mIoU on ADE20K) among models under one billion parameters. We will release the code after the blind review.

Effectively encoding multi-scale contextual information is crucial for accurate semantic segmentation. Most of the existing transformer-based segmentation models combine features across scales without any selection, where features on sub-optimal scales may degrade segmentation outcomes. Leveraging from the inherent properties of Vision Transformers, we propose a simple yet effective module, Transformer Scale Gate (TSG), to optimally combine multi-scale features. TSG exploits cues in self and cross attentions in Vision Transformers for the scale selection. TSG is a highly flexible plug-and-play module, and can easily be incorporated with any encoder-decoder-based hierarchical vision Transformer architecture. Extensive experiments on the Pascal Context, ADE20K and Cityscapes datasets demonstrate that our feature selection strategy achieves consistent gains.

Existing semantic segmentation methods improve generalization capability, by regularizing various images to a canonical feature space. While this process contributes to generalization, it weakens the representation inevitably. In contrast to existing methods, we instead utilize the difference between images to build a better representation space, where the distinct style features are extracted and stored as the bases of representation. Then, the generalization to unseen image styles is achieved by projecting features to this known space. Specifically, we realize the style projection as a weighted combination of stored bases, where the similarity distances are adopted as the weighting factors. Based on the same concept, we extend this process to the decision part of model and promote the generalization of semantic prediction. By measuring the similarity distances to semantic bases (i.e., prototypes), we replace the common deterministic prediction with semantic clustering. Comprehensive experiments demonstrate the advantage of proposed method to the state of the art, up to 3.6% mIoU improvement in average on unseen scenarios.

The existing few-shot medical segmentation networks share the same practice that the more prototypes, the better performance. This phenomenon can be theoretically interpreted in Vector Quantization (VQ) view: the more prototypes, the more clusters are separated from pixel-wise feature points distributed over the full space. However, as we further think about few-shot segmentation with this perspective, it is found that the clusterization of feature points and the adaptation to unseen tasks have not received enough attention. Motivated by the observation, we propose a learning VQ mechanism consisting of grid-format VQ (GFVQ), self-organized VQ (SOVQ) and residual oriented VQ (ROVQ). To be specific, GFVQ generates the prototype matrix by averaging square grids over the spatial extent, which uniformly quantizes the local details; SOVQ adaptively assigns the feature points to different local classes and creates a new representation space where the learnable local prototypes are updated with a global view; ROVQ introduces residual information to fine-tune the aforementioned learned local prototypes without re-training, which benefits the generalization performance for the irrelevance to the training task. We empirically show that our VQ framework yields the state-of-the-art performance over abdomen, cardiac and prostate MRI datasets and expect this work will provoke a rethink of …

Continual Semantic Segmentation (CSS) extends static semantic segmentation by incrementally introducing new classes for training. To alleviate the catastrophic forgetting issue in CSS, a memory buffer that stores a small number of samples from the previous classes is constructed for replay. However, existing methods select the memory samples either randomly or based on a single-factor-driven hand-crafted strategy, which has no guarantee to be optimal. In this work, we propose a novel memory sample selection mechanism that selects informative samples for effective replay in a fully automatic way by considering comprehensive factors including sample diversity and class performance. Our mechanism regards the selection operation as a decision-making process and learns an optimal selection policy that directly maximizes the validation performance on a reward set. To facilitate the selection decision, we design a novel state representation and a dual-stage action space. Our extensive experiments on Pascal-VOC 2012 and ADE 20K datasets demonstrate the effectiveness of our approach with state-of-the-art (SOTA) performance achieved, outperforming the second-place one by 12.54% for the 6-stage setting on Pascal-VOC 2012.

Weakly-Supervised Semantic Segmentation (WSSS) using image-level labels typically utilizes Class Activation Map (CAM) to generate the pseudo labels. Limited by the local structure perception of CNN, CAM usually cannot identify the integral object regions. Though the recent Vision Transformer (ViT) can remedy this flaw, we observe it also brings the over-smoothing issue, ie, the final patch tokens incline to be uniform. In this work, we propose Token Contrast (ToCo) to address this issue and further explore the virtue of ViT for WSSS. Firstly, motivated by the observation that intermediate layers in ViT can still retain semantic diversity, we designed a Patch Token Contrast module (PTC). PTC supervises the final patch tokens with the pseudo token relations derived from intermediate layers, allowing them to align the semantic regions and thus yield more accurate CAM. Secondly, to further differentiate the low-confidence regions in CAM, we devised a Class Token Contrast module (CTC) inspired by the fact that class tokens in ViT can capture high-level semantics. CTC facilitates the representation consistency between uncertain local regions and global objects by contrasting their class tokens. Experiments on the PASCAL VOC and MS COCO datasets show the proposed ToCo can remarkably surpass other single-stage competitors and …

The archaeological dating of bronze dings has played a critical role in the study of ancient Chinese history. Current archaeology depends on trained experts to carry out bronze dating, which is time-consuming and labor-intensive. For such dating, in this study, we propose a learning-based approach to integrate advanced deep learning techniques and archaeological knowledge. To achieve this, we first collect a large-scale image dataset of bronze dings, which contains richer attribute information than other existing fine-grained datasets. Second, we introduce a multihead classifier and a knowledge-guided relation graph to mine the relationship between attributes and the ding era. Third, we conduct comparison experiments with various existing methods, the results of which show that our dating method achieves a state-of-the-art performance. We hope that our data and applied networks will enrich fine-grained classification research relevant to other interdisciplinary areas of expertise. The dataset and source code used are included in our supplementary materials, and will be open after submission owing to the anonymity policy. Source codes and data are available at: https://github.com/zhourixin/bronze-Ding.

Recent semi-supervised semantic segmentation methods combine pseudo labeling and consistency regularization to enhance model generalization from perturbation-invariant training. In this work, we argue that adequate supervision can be extracted directly from the geometry of feature space. Inspired by density-based unsupervised clustering, we propose to leverage feature density to locate sparse regions within feature clusters defined by label and pseudo labels. The hypothesis is that lower-density features tend to be under-trained compared with those densely gathered. Therefore, we propose to apply regularization on the structure of the cluster by tackling the sparsity to increase intra-class compactness in feature space. With this goal, we present a Density-Guided Contrastive Learning (DGCL) strategy to push anchor features in sparse regions toward cluster centers approximated by high-density positive keys. The heart of our method is to estimate feature density which is defined as neighbor compactness. We design a multi-scale density estimation module to obtain the density from multiple nearest-neighbor graphs for robust density modeling. Moreover, a unified training framework is proposed to combine label-guided self-training and density-guided geometry regularization to form complementary supervision on unlabeled data. Experimental results on PASCAL VOC and Cityscapes under various semi-supervised settings demonstrate that our proposed method achieves state-of-the-art performances.

We propose Cut-and-LEaRn (CutLER), a simple approach for training unsupervised object detection and segmentation models. We leverage the property of self-supervised models to ‘discover’ objects without supervision and amplify it to train a state-of-the-art localization model without any human labels. CutLER first uses our proposed MaskCut approach to generate coarse masks for multiple objects in an image, and then learns a detector on these masks using our robust loss function. We further improve performance by self-training the model on its predictions. Compared to prior work, CutLER is simpler, compatible with different detection architectures, and detects multiple objects. CutLER is also a zero-shot unsupervised detector and improves detection performance AP_50 by over 2.7× on 11 benchmarks across domains like video frames, paintings, sketches, etc. With finetuning, CutLER serves as a low-shot detector surpassing MoCo-v2 by 7.3% AP^box and 6.6% AP^mask on COCO when training with 5% labels.

Extracting class activation maps (CAM) from a classification model often results in poor coverage on foreground objects, i.e., only the discriminative region (e.g., the “head” of “sheep”) is recognized and the rest (e.g., the “leg” of “sheep”) mistakenly as background. The crux behind is that the weight of the classifier (used to compute CAM) captures only the discriminative features of objects. We tackle this by introducing a new computation method for CAM that explicitly captures non-discriminative features as well, thereby expanding CAM to cover whole objects. Specifically, we omit the last pooling layer of the classification model, and perform clustering on all local features of an object class, where “local” means “at a spatial pixel position”. We call the resultant K cluster centers local prototypes - represent local semantics like the “head”, “leg”, and “body” of “sheep”. Given a new image of the class, we compare its unpooled features to every prototype, derive K similarity matrices, and then aggregate them into a heatmap (i.e., our CAM). Our CAM thus captures all local features of the class without discrimination. We evaluate it in the challenging tasks of weakly-supervised semantic segmentation (WSSS), and plug it in multiple state-of-the-art WSSS methods, such as …

Labeling objects with pixel-wise segmentation requires a huge amount of human labor compared to bounding boxes. Most existing methods for weakly supervised instance segmentation focus on designing heuristic losses with priors from bounding boxes. While, we find that box-supervised methods can produce some fine segmentation masks and we wonder whether the detectors could learn from these fine masks while ignoring low-quality masks. To answer this question, we present BoxTeacher, an efficient and end-to-end training framework for high-performance weakly supervised instance segmentation, which leverages a sophisticated teacher to generate high-quality masks as pseudo labels. Considering the massive noisy masks hurt the training, we present a mask-aware confidence score to estimate the quality of pseudo masks and propose the noise-aware pixel loss and noise-reduced affinity loss to adaptively optimize the student with pseudo masks. Extensive experiments can demonstrate the effectiveness of the proposed BoxTeacher. Without bells and whistles, BoxTeacher remarkably achieves 35.0 mask AP and 36.5 mask AP with ResNet-50 and ResNet-101 respectively on the challenging COCO dataset, which outperforms the previous state-of-the-art methods by a significant margin and bridges the gap between box-supervised and mask-supervised methods. The code and models will be available later.

Differences in forgery attributes of images generated in CNN-synthesized and image-editing domains are large, and such differences make a unified image forgery detection and localization (IFDL) challenging. To this end, we present a hierarchical fine-grained formulation for IFDL representation learning. Specifically, we first represent forgery attributes of a manipulated image with multiple labels at different levels. Then we perform fine-grained classification at these levels using the hierarchical dependency between them. As a result, the algorithm is encouraged to learn both comprehensive features and inherent hierarchical nature of different forgery attributes, thereby improving the IFDL representation. Our proposed IFDL framework contains three components: multi-branch feature extractor, localization and classification modules. Each branch of the feature extractor learns to classify forgery attributes at one level, while localization and classification modules segment the pixel-level forgery region and detect image-level forgery, respectively. Lastly, we construct a hierarchical fine-grained dataset to facilitate our study. We demonstrate the effectiveness of our method on 7 different benchmarks, for both tasks of IFDL and forgery attribute classification. Our source code and dataset can be found at https://github.com/CHELSEA234/HiFi_IFDL

Image recognition on expert domains is usually fine-grained and requires expert labeling, which is costly. This limits dataset sizes and the accuracy of learning systems. To address this challenge, we consider annotating expert data with crowdsourcing. This is denoted as PrOfeSsional lEvel cRowd (POSER) annotation. A new approach, based on semi-supervised learning (SSL) and denoted as SSL with human filtering (SSL-HF) is proposed. It is a human-in-the-loop SSL method, where crowd-source workers act as filters of pseudo-labels, replacing the unreliable confidence thresholding used by state-of-the-art SSL methods. To enable annotation by non-experts, classes are specified implicitly, via positive and negative sets of examples and augmented with deliberative explanations, which highlight regions of class ambiguity. In this way, SSL-HF leverages the strong low-shot learning and confidence estimation ability of humans to create an intuitive but effective labeling experience. Experiments show that SSL-HF significantly outperforms various alternative approaches in several benchmarks.

Recent advances in self-supervised visual representation learning have paved the way for unsupervised methods tackling tasks such as object discovery and instance segmentation. However, discovering objects in an image with no supervision is a very hard task; what are the desired objects, when to separate them into parts, how many are there, and of what classes? The answers to these questions depend on the tasks and datasets of evaluation. In this work, we take a different approach and propose to look for the background instead. This way, the salient objects emerge as a by-product without any strong assumption on what an object should be. We propose FOUND, a simple model made of a single conv 1x1 initialized with coarse background masks extracted from self-supervised patch-based representations. After fast training and refining these seed masks, the model reaches state-of-the-art results on unsupervised saliency detection and object discovery benchmarks. Moreover, we show that our approach yields good results in the unsupervised semantic segmentation retrieval task. The code to reproduce our results is available at https://github.com/valeoai/FOUND.

Self-supervised learning models have been shown to learn rich visual representations without requiring human annotations. However, in many real-world scenarios, labels are partially available, motivating a recent line of work on semi-supervised methods inspired by self-supervised principles. In this paper, we propose a conceptually simple yet empirically powerful approach to turn clustering-based self-supervised methods such as SwAV or DINO into semi-supervised learners. More precisely, we introduce a multi-task framework merging a supervised objective using ground-truth labels and a self-supervised objective relying on clustering assignments with a single cross-entropy loss. This approach may be interpreted as imposing the cluster centroids to be class prototypes. Despite its simplicity, we provide empirical evidence that our approach is highly effective and achieves state-of-the-art performance on CIFAR100 and ImageNet.

During training, supervised object detection tries to correctly match the predicted bounding boxes and associated classification scores to the ground truth. This is essential to determine which predictions are to be pushed towards which solutions, or to be discarded. Popular matching strategies include matching to the closest ground truth box (mostly used in combination with anchors), or matching via the Hungarian algorithm (mostly used in anchor-free methods). Each of these strategies comes with its own properties, underlying losses, and heuristics. We show how Unbalanced Optimal Transport unifies these different approaches and opens a whole continuum of methods in between. This allows for a finer selection of the desired properties. Experimentally, we show that training an object detection model with Unbalanced Optimal Transport is able to reach the state-of-the-art both in terms of Average Precision and Average Recall as well as to provide a faster initial convergence. The approach is well suited for GPU implementation, which proves to be an advantage for large-scale models.

Generalized few-shot object detection aims to achieve precise detection on both base classes with abundant annotations and novel classes with limited training data. Existing approaches enhance few-shot generalization with the sacrifice of base-class performance, or maintain high precision in base-class detection with limited improvement in novel-class adaptation. In this paper, we point out the reason is insufficient Discriminative feature learning for all of the classes. As such, we propose a new training framework, DiGeo, to learn Geometry-aware features of inter-class separation and intra-class compactness. To guide the separation of feature clusters, we derive an offline simplex equiangular tight frame (ETF) classifier whose weights serve as class centers and are maximally and equally separated. To tighten the cluster for each class, we include adaptive class-specific margins into the classification loss and encourage the features close to the class centers. Experimental studies on two few-shot benchmark datasets (PASCAL VOC, MSCOCO) and one long-tail dataset (LVIS) demonstrate that, with a single model, our method can effectively improve generalization on novel classes without hurting the detection of base classes.

Single Domain Generalization (SDG) tackles the problem of training a model on a single source domain so that it generalizes to any unseen target domain. While this has been well studied for image classification, the literature on SDG object detection remains almost non-existent. To address the challenges of simultaneously learning robust object localization and representation, we propose to leverage a pre-trained vision-language model to introduce semantic domain concepts via textual prompts. We achieve this via a semantic augmentation strategy acting on the features extracted by the detector backbone, as well as a text-based classification loss. Our experiments evidence the benefits of our approach, outperforming by 10% the only existing SDG object detection method, Single-DGOD[49], on their own diverse weather-driving benchmark.

The recently proposed open-world object and open-set detection have achieved a breakthrough in finding never-seen-before objects and distinguishing them from known ones. However, their studies on knowledge transfer from known classes to unknown ones are not deep enough, resulting in the scanty capability for detecting unknowns hidden in the background. In this paper, we propose the unknown sniffer (UnSniffer) to find both unknown and known objects. Firstly, the generalized object confidence (GOC) score is introduced, which only uses known samples for supervision and avoids improper suppression of unknowns in the background. Significantly, such confidence score learned from known objects can be generalized to unknown ones. Additionally, we propose a negative energy suppression loss to further suppress the non-object samples in the background. Next, the best box of each unknown is hard to obtain during inference due to lacking their semantic information in training. To solve this issue, we introduce a graph-based determination scheme to replace hand-designed non-maximum suppression (NMS) post-processing. Finally, we present the Unknown Object Detection Benchmark, the first publicly benchmark that encompasses precision evaluation for unknown detection to our knowledge. Experiments show that our method is far better than the existing state-of-the-art methods. Code is available at: https://github.com/Went-Liang/UnSniffer.

In this study, we dive deep into the inconsistency of pseudo targets in semi-supervised object detection (SSOD). Our core observation is that the oscillating pseudo-targets undermine the training of an accurate detector. It injects noise into the student’s training, leading to severe overfitting problems. Therefore, we propose a systematic solution, termed NAME, to reduce the inconsistency. First, adaptive anchor assignment~(ASA) substitutes the static IoU-based strategy, which enables the student network to be resistant to noisy pseudo-bounding boxes. Then we calibrate the subtask predictions by designing a 3D feature alignment module~(FAM-3D). It allows each classification feature to adaptively query the optimal feature vector for the regression task at arbitrary scales and locations. Lastly, a Gaussian Mixture Model (GMM) dynamically revises the score threshold of pseudo-bboxes, which stabilizes the number of ground truths at an early stage and remedies the unreliable supervision signal during training. NAME provides strong results on a large range of SSOD evaluations. It achieves 40.0 mAP with ResNet-50 backbone given only 10% of annotated MS-COCO data, which surpasses previous baselines using pseudo labels by around 3 mAP. When trained on fully annotated MS-COCO with additional unlabeled data, the performance further increases to 47.7 mAP. Our code is available …

End-to-end pedestrian detection focuses on training a pedestrian detection model via discarding the Non-Maximum Suppression (NMS) post-processing. Though a few methods have been explored, most of them still suffer from longer training time and more complex deployment, which cannot be deployed in the actual industrial applications. In this paper, we intend to bridge this gap and propose an Optimal Proposal Learning (OPL) framework for deployable end-to-end pedestrian detection. Specifically, we achieve this goal by using CNN-based light detector and introducing two novel modules, including a Coarse-to-Fine (C2F) learning strategy for proposing precise positive proposals for the Ground-Truth (GT) instances by reducing the ambiguity of sample assignment/output in training/testing respectively, and a Completed Proposal Network (CPN) for producing extra information compensation to further recall the hard pedestrian samples. Extensive experiments are conducted on CrowdHuman, TJU-Ped and Caltech, and the results show that our proposed OPL method significantly outperforms the competing methods.

In this work, we study few-shot domain adaptive object detection (FSDAOD), where only a few target labeled images are available for training in addition to sufficient source labeled images. Critically, in FSDAOD, the data-scarcity in the target domain leads to an extreme data imbalance between the source and target domains, which potentially causes over-adaptation in traditional feature alignment. To address the data imbalance problem, we propose an asymmetric adaptation paradigm, namely AsyFOD, which leverages the source and target instances from different perspectives. Specifically, by using target distribution estimation, the AsyFOD first identifies the target-similar source instances, which serves for augmenting the limited target instances. Then, we conduct asynchronous alignment between target-dissimilar source instances and augmented target instances, which is simple yet effective for alleviating the over-adaptation. Extensive experiments demonstrate that the proposed AsyFOD outperforms all state-of-the-art methods on four FSDAOD benchmarks with various environmental variances, e.g., 3.1% mAP improvement on Cityscapes-to-FoggyCityscapes and 2.9% mAP increase on Sim10k-to-Cityscapes. The code is available at https://github.com/Hlings/AsyFOD.

Image generation relies on massive training data that can hardly produce diverse images of an unseen category according to a few examples. In this paper, we address this dilemma by projecting sparse few-shot samples into a continuous latent space that can potentially generate infinite unseen samples. The rationale behind is that we aim to locate a centroid latent position in a conditional StyleGAN, where the corresponding output image on that centroid can maximize the similarity with the given samples. Although the given samples are unseen for the conditional StyleGAN, we assume the neighboring latent subspace around the centroid belongs to the novel category, and therefore introduce two latent subspace optimization objectives. In the first one we use few-shot samples as positive anchors of the novel class, and adjust the StyleGAN to produce the corresponding results with the new class label condition. The second objective is to govern the generation process from the other way around, by altering the centroid and its surrounding latent subspace for a more precise generation of the novel class. These reciprocal optimization objectives inject a novel class into the StyleGAN latent subspace, and therefore new unseen samples can be easily produced by sampling images from it. …

Semantically coherent out-of-distribution (SCOOD) detection aims to discern outliers from the intended data distribution with access to unlabeled extra set. The coexistence of in-distribution and out-of-distribution samples will exacerbate the model overfitting when no distinction is made. To address this problem, we propose a novel uncertainty-aware optimal transport scheme. Our scheme consists of an energy-based transport (ET) mechanism that estimates the fluctuating cost of uncertainty to promote the assignment of semantic-agnostic representation, and an inter-cluster extension strategy that enhances the discrimination of semantic property among different clusters by widening the corresponding margin distance. Furthermore, a T-energy score is presented to mitigate the magnitude gap between the parallel transport and classifier branches. Extensive experiments on two standard SCOOD benchmarks demonstrate the above-par OOD detection performance, outperforming the state-of-the-art methods by a margin of 27.69% and 34.4% on FPR@95, respectively.

Accurate segmentation of cellular images remains an elusive task due to the intrinsic variability in morphology of biological structures. Complete manual segmentation is unfeasible for large datasets, and while supervised methods have been proposed to automate segmentation, they often rely on manually generated ground truths which are especially challenging and time consuming to generate in biology due to the requirement of domain expertise. Furthermore, these methods have limited generalization capacity, requiring additional manual labels to be generated for each dataset and use case. We introduce MAESTER (Masked AutoEncoder guided SegmenTation at pixEl Resolution), a self-supervised method for accurate, subcellular structure segmentation at pixel resolution. MAESTER treats segmentation as a representation learning and clustering problem. Specifically, MAESTER learns semantically meaningful token representations of multi-pixel image patches while simultaneously maintaining a sufficiently large field of view for contextual learning. We also develop a cover-and-stride inference strategy to achieve pixel-level subcellular structure segmentation. We evaluated MAESTER on a publicly available volumetric electron microscopy (VEM) dataset of primary mouse pancreatic islets beta cells and achieved upwards of 29.1% improvement over state-of-the-art under the same evaluation criteria. Furthermore, our results are competitive against supervised methods trained on the same tasks, closing the gap between self-supervised …

Recent trends in semi-supervised learning have significantly boosted the performance of 3D semi-supervised medical image segmentation. Compared with 2D images, 3D medical volumes involve information from different directions, e.g., transverse, sagittal, and coronal planes, so as to naturally provide complementary views. These complementary views and the intrinsic similarity among adjacent 3D slices inspire us to develop a novel annotation way and its corresponding semi-supervised model for effective segmentation. Specifically, we firstly propose the orthogonal annotation by only labeling two orthogonal slices in a labeled volume, which significantly relieves the burden of annotation. Then, we perform registration to obtain the initial pseudo labels for sparsely labeled volumes. Subsequently, by introducing unlabeled volumes, we propose a dual-network paradigm named Dense-Sparse Co-training (DeSCO) that exploits dense pseudo labels in early stage and sparse labels in later stage and meanwhile forces consistent output of two networks. Experimental results on three benchmark datasets validated our effectiveness in performance and efficiency in annotation. For example, with only 10 annotated slices, our method reaches a Dice up to 86.93% on KiTS19 dataset.

In biological research, fluorescence staining is a key technique to reveal the locations and morphology of subcellular structures. However, it is slow, expensive, and harmful to cells. In this paper, we model it as a deep learning task termed subcellular structure prediction (SSP), aiming to predict the 3D fluorescent images of multiple subcellular structures from a 3D transmitted-light image. Unfortunately, due to the limitations of current biotechnology, each image is partially labeled in SSP. Besides, naturally, subcellular structures vary considerably in size, which causes the multi-scale issue of SSP. To overcome these challenges, we propose Re-parameterizing Mixture-of-Diverse-Experts (RepMode), a network that dynamically organizes its parameters with task-aware priors to handle specified single-label prediction tasks. In RepMode, the Mixture-of-Diverse-Experts (MoDE) block is designed to learn the generalized parameters for all tasks, and gating re-parameterization (GatRep) is performed to generate the specialized parameters for each task, by which RepMode can maintain a compact practical topology exactly like a plain network, and meanwhile achieves a powerful theoretical topology. Comprehensive experiments show that RepMode can achieve state-of-the-art overall performance in SSP.

In digital pathology, the spatial context of cells is important for cell classification, cancer diagnosis and prognosis. To model such complex cell context, however, is challenging. Cells form different mixtures, lineages, clusters and holes. To model such structural patterns in a learnable fashion, we introduce several mathematical tools from spatial statistics and topological data analysis. We incorporate such structural descriptors into a deep generative model as both conditional inputs and a differentiable loss. This way, we are able to generate high quality multi-class cell layouts for the first time. We show that the topology-rich cell layouts can be used for data augmentation and improve the performance of downstream tasks such as cell classification.

Automatic radiology reporting has great clinical potential to relieve radiologists from heavy workloads and improve diagnosis interpretation. Recently, researchers have enhanced data-driven neural networks with medical knowledge graphs to eliminate the severe visual and textual bias in this task. The structures of such graphs are exploited by using the clinical dependencies formed by the disease topic tags via general knowledge and usually do not update during the training process. Consequently, the fixed graphs can not guarantee the most appropriate scope of knowledge and limit the effectiveness. To address the limitation, we propose a knowledge graph with Dynamic structure and nodes to facilitate chest X-ray report generation with Contrastive Learning, named DCL. In detail, the fundamental structure of our graph is pre-constructed from general knowledge. Then we explore specific knowledge extracted from the retrieved reports to add additional nodes or redefine their relations in a bottom-up manner. Each image feature is integrated with its very own updated graph before being fed into the decoder module for report generation. Finally, this paper introduces Image-Report Contrastive and Image-Report Matching losses to better represent visual features and textual information. Evaluated on IU-Xray and MIMIC-CXR datasets, our DCL outperforms previous state-of-the-art models on these two …

Computational pathology can lead to saving human lives, but models are annotation hungry and pathology images are notoriously expensive to annotate. Self-supervised learning has shown to be an effective method for utilizing unlabeled data, and its application to pathology could greatly benefit its downstream tasks. Yet, there are no principled studies that compare SSL methods and discuss how to adapt them for pathology. To address this need, we execute the largest-scale study of SSL pre-training on pathology image data, to date. Our study is conducted using 4 representative SSL methods on diverse downstream tasks. We establish that large-scale domain-aligned pre-training in pathology consistently out-performs ImageNet pre-training in standard SSL settings such as linear and fine-tuning evaluations, as well as in low-label regimes. Moreover, we propose a set of domain-specific techniques that we experimentally show leads to a performance boost. Lastly, for the first time, we apply SSL to the challenging task of nuclei instance segmentation and show large and consistent performance improvements under diverse settings.

Learning representations for individual instances when only bag-level labels are available is a fundamental challenge in multiple instance learning (MIL). Recent works have shown promising results using contrastive self-supervised learning (CSSL), which learns to push apart representations corresponding to two different randomly-selected instances. Unfortunately, in real-world applications such as medical image classification, there is often class imbalance, so randomly-selected instances mostly belong to the same majority class, which precludes CSSL from learning inter-class differences. To address this issue, we propose a novel framework, Iterative Self-paced Supervised Contrastive Learning for MIL Representations (ItS2CLR), which improves the learned representation by exploiting instance-level pseudo labels derived from the bag-level labels. The framework employs a novel self-paced sampling strategy to ensure the accuracy of pseudo labels. We evaluate ItS2CLR on three medical datasets, showing that it improves the quality of instance-level pseudo labels and representations, and outperforms existing MIL methods in terms of both bag and instance level accuracy. Code is available at https://github.com/Kangningthu/ItS2CLR

Prompt learning is an efficient approach to adapt transformers by inserting learnable set of parameters into the input and intermediate representations of a pre-trained model. In this work, we present Expressive Prompts with Residuals (EXPRES) which modifies the prompt learning paradigm specifically for effective adaptation of vision transformers (ViT). Out method constructs downstream representations via learnable “output” tokens, that are akin to the learned class tokens of the ViT. Further for better steering of the downstream representation processed by the frozen transformer, we introduce residual learnable tokens that are added to the output of various computations. We apply EXPRES for image classification, few shot learning, and semantic segmentation, and show our method is capable of achieving state of the art prompt tuning on 3/3 categories of the VTAB benchmark. In addition to strong performance, we observe that our approach is an order of magnitude more prompt efficient than existing visual prompting baselines. We analytically show the computational benefits of our approach over weight space adaptation techniques like finetuning. Lastly we systematically corroborate the architectural design of our method via a series of ablation experiments.

In machine learning, it is often observed that standard training outputs anomalously high confidence for both in-distribution (ID) and out-of-distribution (OOD) data. Thus, the ability to detect OOD samples is critical to the model deployment. An essential step for OOD detection is post-hoc scoring. MaxLogit is one of the simplest scoring functions which uses the maximum logits as OOD score. To provide a new viewpoint to study the logit-based scoring function, we reformulate the logit into cosine similarity and logit norm and propose to use MaxCosine and MaxNorm. We empirically find that MaxCosine is a core factor in the effectiveness of MaxLogit. And the performance of MaxLogit is encumbered by MaxNorm. To tackle the problem, we propose the Decoupling MaxLogit (DML) for flexibility to balance MaxCosine and MaxNorm. To further embody the core of our method, we extend DML to DML+ based on the new insights that fewer hard samples and compact feature space are the key components to make logit-based methods effective. We demonstrate the effectiveness of our logit-based OOD detection methods on CIFAR-10, CIFAR-100 and ImageNet and establish state-of-the-art performance.

Multi-label recognition (MLR) with incomplete labels is very challenging. Recent works strive to explore the image-to-label correspondence in the vision-language model, i.e., CLIP, to compensate for insufficient annotations. In spite of promising performance, they generally overlook the valuable prior about the label-to-label correspondence. In this paper, we advocate remedying the deficiency of label supervision for the MLR with incomplete labels by deriving a structured semantic prior about the label-to-label correspondence via a semantic prior prompter. We then present a novel Semantic Correspondence Prompt Network (SCPNet), which can thoroughly explore the structured semantic prior. A Prior-Enhanced Self-Supervised Learning method is further introduced to enhance the use of the prior. Comprehensive experiments and analyses on several widely used benchmark datasets show that our method significantly outperforms existing methods on all datasets, well demonstrating the effectiveness and the superiority of our method. Our code will be available at https://github.com/jameslahm/SCPNet.

Due to the expensive costs of collecting labels in multi-label classification datasets, partially annotated multi-label classification has become an emerging field in computer vision. One baseline approach to this task is to assume unobserved labels as negative labels, but this assumption induces label noise as a form of false negative. To understand the negative impact caused by false negative labels, we study how these labels affect the model’s explanation. We observe that the explanation of two models, trained with full and partial labels each, highlights similar regions but with different scaling, where the latter tends to have lower attribution scores. Based on these findings, we propose to boost the attribution scores of the model trained with partial labels to make its explanation resemble that of the model trained with full labels. Even with the conceptually simple approach, the multi-label classification performance improves by a large margin in three different datasets on a single positive label setting and one on a large-scale partial label setting. Code is available at https://github.com/youngwk/BridgeGapExplanationPAMC.

Clustering has been a major research topic in the field of machine learning, one to which Deep Learning has recently been applied with significant success. However, an aspect of clustering that is not addressed by existing deep clustering methods, is that of efficiently producing multiple, diverse partitionings for a given dataset. This is particularly important, as a diverse set of base clusterings are necessary for consensus clustering, which has been found to produce better and more robust results than relying on a single clustering. To address this gap, we propose DivClust, a diversity controlling loss that can be incorporated into existing deep clustering frameworks to produce multiple clusterings with the desired degree of diversity. We conduct experiments with multiple datasets and deep clustering frameworks and show that: a) our method effectively controls diversity across frameworks and datasets with very small additional computational cost, b) the sets of clusterings learned by DivClust include solutions that significantly outperform single-clustering baselines, and c) using an off-the-shelf consensus clustering algorithm, DivClust produces consensus clustering solutions that consistently outperform single-clustering baselines, effectively improving the performance of the base deep clustering framework.

Metric learning requires the identification of far-apart similar pairs and close dissimilar pairs during training, and this is difficult to achieve with unlabeled data because pairs are typically assumed to be similar if they are close. We present a novel metric learning method which circumvents this issue by identifying hard negative pairs as those which obtain dissimilar labels via label propagation (LP), when the edge linking the pair of data is removed in the affinity matrix. In so doing, the negative pairs can be identified despite their proximity, and we are able to utilize this information to significantly improve LP’s ability to identify far-apart positive pairs and close negative pairs. This results in a considerable improvement in semi-supervised metric learning performance as evidenced by recall, precision and Normalized Mutual Information (NMI) performance metrics on Content-based Information Retrieval (CBIR) applications.

In practical settings, classification datasets are obtained through a labelling process that is usually done by humans. Labels can be noisy as they are obtained by aggregating the different individual labels assigned to the same sample by multiple, and possibly disagreeing, annotators. The inter-rater agreement on these datasets can be measured while the underlying noise distribution to which the labels are subject is assumed to be unknown. In this work, we: (i) show how to leverage the inter-annotator statistics to estimate the noise distribution to which labels are subject; (ii) introduce methods that use the estimate of the noise distribution to learn from the noisy dataset; and (iii) establish generalization bounds in the empirical risk minimization framework that depend on the estimated quantities. We conclude the paper by providing experiments that illustrate our findings.

Novel class discovery (NCD) aims at learning a model that transfers the common knowledge from a class-disjoint labelled dataset to another unlabelled dataset and discovers new classes (clusters) within it. Many methods, as well as elaborate training pipelines and appropriate objectives, have been proposed and considerably boosted performance on NCD tasks. Despite all this, we find that the existing methods do not sufficiently take advantage of the essence of the NCD setting. To this end, in this paper, we propose to model both inter-class and intra-class constraints in NCD based on the symmetric Kullback-Leibler divergence (sKLD). Specifically, we propose an inter-class sKLD constraint to effectively exploit the disjoint relationship between labelled and unlabelled classes, enforcing the separability for different classes in the embedding space. In addition, we present an intra-class sKLD constraint to explicitly constrain the intra-relationship between a sample and its augmentations and ensure the stability of the training process at the same time. We conduct extensive experiments on the popular CIFAR10, CIFAR100 and ImageNet benchmarks and successfully demonstrate that our method can establish a new state of the art and can achieve significant performance improvements, e.g., 3.5%/3.7% clustering accuracy improvements on CIFAR100-50 dataset split under the task-aware/-agnostic evaluation …

Novel Class Discovery (NCD) aims to discover unknown classes without any annotation, by exploiting the transferable knowledge already learned from a base set of known classes. Existing works hold an impractical assumption that the novel class distribution prior is uniform, yet neglect the imbalanced nature of real-world data. In this paper, we relax this assumption by proposing a new challenging task: distribution-agnostic NCD, which allows data drawn from arbitrary unknown class distributions and thus renders existing methods useless or even harmful. We tackle this challenge by proposing a new method, dubbed “Bootstrapping Your Own Prior (BYOP)”, which iteratively estimates the class prior based on the model prediction itself. At each iteration, we devise a dynamic temperature technique that better estimates the class prior by encouraging sharper predictions for less-confident samples. Thus, BYOP obtains more accurate pseudo-labels for the novel samples, which are beneficial for the next training iteration. Extensive experiments show that existing methods suffer from imbalanced class distributions, while BYOP outperforms them by clear margins, demonstrating its effectiveness across various distribution scenarios.

While long-tailed semi-supervised learning (LTSSL) has received tremendous attention in many real-world classification problems, existing LTSSL algorithms typically assume that the class distributions of labeled and unlabeled data are almost identical. Those LTSSL algorithms built upon the assumption can severely suffer when the class distributions of labeled and unlabeled data are mismatched since they utilize biased pseudo-labels from the model. To alleviate this issue, we propose a new simple method that can effectively utilize unlabeled data of unknown class distributions by introducing the adaptive consistency regularizer (ACR). ACR realizes the dynamic refinery of pseudo-labels for various distributions in a unified formula by estimating the true class distribution of unlabeled data. Despite its simplicity, we show that ACR achieves state-of-the-art performance on a variety of standard LTSSL benchmarks, e.g., an averaged 10% absolute increase of test accuracy against existing algorithms when the class distributions of labeled and unlabeled data are mismatched. Even when the class distributions are identical, ACR consistently outperforms many sophisticated LTSSL algorithms. We carry out extensive ablation studies to tease apart the factors that are most important to ACR’s success. Source code is available at https://github.com/Gank0078/ACR.

Although existing semi-supervised learning models achieve remarkable success in learning with unannotated in-distribution data, they mostly fail to learn on unlabeled data sampled from novel semantic classes due to their closed-set assumption. In this work, we target a pragmatic but under-explored Generalized Novel Category Discovery (GNCD) setting. The GNCD setting aims to categorize unlabeled training data coming from known and novel classes by leveraging the information of partially labeled known classes. We propose a two-stage Contrastive Affinity Learning method with auxiliary visual Prompts, dubbed PromptCAL, to address this challenging problem. Our approach discovers reliable pairwise sample affinities to learn better semantic clustering of both known and novel classes for the class token and visual prompts. First, we propose a discriminative prompt regularization loss to reinforce semantic discriminativeness of prompt-adapted pre-trained vision transformer for refined affinity relationships. Besides, we propose contrastive affinity learning to calibrate semantic representations based on our iterative semi-supervised affinity graph generation method for semantically-enhanced supervision. Extensive experimental evaluation demonstrates that our PromptCAL method is more effective in discovering novel classes even with limited annotations and surpasses the current state-of-the-art on generic and fine-grained benchmarks (e.g., with nearly 11% gain on CUB-200, and 9% on ImageNet-100) on overall …

Mean ensemble (i.e. averaging predictions from multiple models) is a commonly-used technique in machine learning that improves the performance of each individual model. We formalize it as feature alignment for ensemble in open-set face recognition and generalize it into Bayesian Ensemble Averaging (BEA) through the lens of probabilistic modeling. This generalization brings up two practical benefits that existing methods could not provide: (1) the uncertainty of a face image can be evaluated and further decomposed into aleatoric uncertainty and epistemic uncertainty, the latter of which can be used as a measure for out-of-distribution detection of faceness; (2) a BEA statistic provably reflects the aleatoric uncertainty of a face image, acting as a measure for face image quality to improve recognition performance. To inherit the uncertainty estimation capability from BEA without the loss of inference efficiency, we propose BEA-KD, a student model to distill knowledge from BEA. BEA-KD mimics the overall behavior of ensemble members and consistently outperforms SOTA knowledge distillation methods on various challenging benchmarks.

It’s widely acknowledged that deep learning models with flatter minima in its loss landscape tend to generalize better. However, such property is under-explored in deep long-tailed recognition (DLTR), a practical problem where the model is required to generalize equally well across all classes when trained on highly imbalanced label distribution. In this paper, through empirical observations, we argue that sharp minima are in fact prevalent in deep longtailed models, whereas naïve integration of existing flattening operations into long-tailed learning algorithms brings little improvement. Instead, we propose an effective twostage sharpness-aware optimization approach based on the decoupling paradigm in DLTR. In the first stage, both the feature extractor and classifier are trained under parameter perturbations at a class-conditioned scale, which is theoretically motivated by the characteristic radius of flat minima under the PAC-Bayesian framework. In the second stage, we generate adversarial features with classbalanced sampling to further robustify the classifier with the backbone frozen. Extensive experiments on multiple longtailed visual recognition benchmarks show that, our proposed Class-Conditional Sharpness-Aware Minimization (CC-SAM), achieves competitive performance compared to the state-of-the-arts. Code is available at https:// github.com/zzpustc/CC-SAM.

Domain Generalization (DG) has achieved great success in generalizing knowledge from source domains to unseen target domains. However, current DG methods rely heavily on labeled source data, which are usually costly and unavailable. Since unlabeled data are far more accessible, we study a more practical unsupervised domain generalization (UDG) problem. Learning invariant visual representation from different views, i.e., contrastive learning, promises well semantic features for in-domain unsupervised learning. However, it fails in cross-domain scenarios. In this paper, we first delve into the failure of vanilla contrastive learning and point out that semantic connectivity is the key to UDG. Specifically, suppressing the intra-domain connectivity and encouraging the intra-class connectivity help to learn the domain-invariant semantic information. Then, we propose a novel unsupervised domain generalization approach, namely Dual Nearest Neighbors contrastive learning with strong Augmentation (DN^2A). Our DN^2A leverages strong augmentations to suppress the intra-domain connectivity and proposes a novel dual nearest neighbors search strategy to find trustworthy cross domain neighbors along with in-domain neighbors to encourage the intra-class connectivity. Experimental results demonstrate that our DN^2A outperforms the state-of-the-art by a large margin, e.g., 12.01% and 13.11% accuracy gain with only 1% labels for linear evaluation on PACS and DomainNet, respectively.

Unsupervised Domain Adaptation (UDA) is an effective approach to tackle the issue of domain shift. Specifically, UDA methods try to align the source and target representations to improve generalization on the target domain. Further, UDA methods work under the assumption that the source data is accessible during the adaptation process. However, in real-world scenarios, the labelled source data is often restricted due to privacy regulations, data transmission constraints, or proprietary data concerns. The Source-Free Domain Adaptation (SFDA) setting aims to alleviate these concerns by adapting a source-trained model for the target domain without requiring access to the source data. In this paper, we explore the SFDA setting for the task of adaptive object detection. To this end, we propose a novel training strategy for adapting a source-trained object detector to the target domain without source data. More precisely, we design a novel contrastive loss to enhance the target representations by exploiting the objects relations for a given target domain input. These object instance relations are modelled using an Instance Relation Graph (IRG) network, which are then used to guide the contrastive representation learning. In addition, we utilize a student-teacher to effectively distill knowledge from source-trained model to target domain. Extensive …

As an important methodology to measure distribution discrepancy, optimal transport (OT) has been successfully applied to learn generalizable visual models under changing environments. However, there are still limitations, including strict prior assumption and implicit alignment, for current OT modeling in challenging real-world scenarios like partial domain adaptation, where the learned transport plan may be biased and negative transfer is inevitable. Thus, it is necessary to explore a more feasible OT methodology for real-world applications. In this work, we focus on the rigorous OT modeling for conditional distribution matching and label shift correction. A novel masked OT (MOT) methodology on conditional distributions is proposed by defining a mask operation with label information. Further, a relaxed and reweighting formulation is proposed to improve the robustness of OT in extreme scenarios. We prove the theoretical equivalence between conditional OT and MOT, which implies the well-defined MOT serves as a computation-friendly proxy. Extensive experiments validate the effectiveness of theoretical results and proposed model.

Visible-infrared recognition (VI recognition) is a challenging task due to the enormous visual difference across heterogeneous images. Most existing works achieve promising results by transfer learning, such as pretraining on the ImageNet, based on advanced neural architectures like ResNet and ViT. However, such methods ignore the negative influence of the pretrained colour prior knowledge, as well as their heavy computational burden makes them hard to deploy in actual scenarios with limited resources. In this paper, we propose a novel task-oriented pretrained lightweight neural network (TOPLight) for VI recognition. Specifically, the TOPLight method simulates the domain conflict and sample variations with the proposed fake domain loss in the pretraining stage, which guides the network to learn how to handle those difficulties, such that a more general modality-shared feature representation is learned for the heterogeneous images. Moreover, an effective fine-grained dependency reconstruction module (FDR) is developed to discover substantial pattern dependencies shared in two modalities. Extensive experiments on VI person re-identification and VI face recognition datasets demonstrate the superiority of the proposed TOPLight, which significantly outperforms the current state of the arts while demanding fewer computational resources.

Extending from unimodal to multimodal is a critical challenge for unsupervised domain adaptation (UDA). Two major problems emerge in unsupervised multimodal domain adaptation: domain adaptation and modality alignment. An intuitive way to handle these two problems is to fulfill these tasks in two separate stages: aligning modalities followed by domain adaptation, or vice versa. However, domains and modalities are not associated in most existing two-stage studies, and the relationship between them is not leveraged which can provide complementary information to each other. In this paper, we unify these two stages into one to align domains and modalities simultaneously. In our model, a tensor-based alignment module (TAL) is presented to explore the relationship between domains and modalities. By this means, domains and modalities can interact sufficiently and guide them to utilize complementary information for better results. Furthermore, to establish a bridge between domains, a dynamic domain generator (DDG) module is proposed to build transitional samples by mixing the shared information of two domains in a self-supervised manner, which helps our model learn a domain-invariant common representation space. Extensive experiments prove that our method can achieve superior performance in two real-world applications. The code will be publicly available.

Endeavors have been recently made to leverage the vision transformer (ViT) for the challenging unsupervised domain adaptation (UDA) task. They typically adopt the cross-attention in ViT for direct domain alignment. However, as the performance of cross-attention highly relies on the quality of pseudo labels for targeted samples, it becomes less effective when the domain gap becomes large. We solve this problem from a game theory’s perspective with the proposed model dubbed as PMTrans, which bridges source and target domains with an intermediate domain. Specifically, we propose a novel ViT-based module called PatchMix that effectively builds up the intermediate domain, i.e., probability distribution, by learning to sample patches from both domains based on the game-theoretical models. This way, it learns to mix the patches from the source and target domains to maximize the cross entropy (CE), while exploiting two semi-supervised mixup losses in the feature and label spaces to minimize it. As such, we interpret the process of UDA as a min-max CE game with three players, including the feature extractor, classifier, and PatchMix, to find the Nash Equilibria. Moreover, we leverage attention maps from ViT to re-weight the label of each patch by its importance, making it possible to obtain …

We introduce anchored radial observations (ARO), a novel shape encoding for learning implicit field representation of 3D shapes that is category-agnostic and generalizable amid significant shape variations. The main idea behind our work is to reason about shapes through partial observations from a set of viewpoints, called anchors. We develop a general and unified shape representation by employing a fixed set of anchors, via Fibonacci sampling, and designing a coordinate-based deep neural network to predict the occupancy value of a query point in space. Differently from prior neural implicit models that use global shape feature, our shape encoder operates on contextual, query-specific features. To predict point occupancy, locally observed shape information from the perspective of the anchors surrounding the input query point are encoded and aggregated through an attention module, before implicit decoding is performed. We demonstrate the quality and generality of our network, coined ARO-Net, on surface reconstruction from sparse point clouds, with tests on novel and unseen object categories, “one-shape” training, and comparisons to state-of-the-art neural and classical methods for reconstruction and tessellation.

Test-time adaptation (TTA) is the problem of updating a pre-trained source model at inference time given test input(s) from a different target domain. Most existing TTA approaches assume the setting in which the target domain is stationary, i.e., all the test inputs come from a single target domain. However, in many practical settings, the test input distribution might exhibit a lifelong/continual shift over time. Moreover, existing TTA approaches also lack the ability to provide reliable uncertainty estimates, which is crucial when distribution shifts occur between the source and target domain. To address these issues, we present PETAL (Probabilistic lifElong Test-time Adaptation with seLf-training prior), which solves lifelong TTA using a probabilistic approach, and naturally results in (1) a student-teacher framework, where the teacher model is an exponential moving average of the student model, and (2) regularizing the model updates at inference time using the source model as a regularizer. To prevent model drift in the lifelong/continual TTA setting, we also propose a data-driven parameter restoration technique which contributes to reducing the error accumulation and maintaining the knowledge of recent domains by restoring only the irrelevant parameters. In terms of predictive error rate as well as uncertainty based metrics such as …

Machine learning society has witnessed the emergence of a myriad of Out-of-Distribution (OoD) algorithms, which address the distribution shift between the training and the testing distribution by searching for a unified predictor or invariant feature representation. However, the task of directly mitigating the distribution shift in the unseen testing set is rarely investigated, due to the unavailability of the testing distribution during the training phase and thus the impossibility of training a distribution translator mapping between the training and testing distribution. In this paper, we explore how to bypass the requirement of testing distribution for distribution translator training and make the distribution translation useful for OoD prediction. We propose a portable Distribution Shift Inversion (DSI) algorithm, in which, before being fed into the prediction model, the OoD testing samples are first linearly combined with additional Gaussian noise and then transferred back towards the training distribution using a diffusion model trained only on the source distribution. Theoretical analysis reveals the feasibility of our method. Experimental results, on both multiple-domain generalization datasets and single-domain generalization datasets, show that our method provides a general performance gain when plugged into a wide range of commonly used OoD algorithms. Our code is available at https://github.com/yu-rp/Distribution-Shift-Iverson}{https://github.com/yu-rp/Distribution-Shift-Iverson.

This paper studies the fundamental problem of learning multi-layer generator models. The multi-layer generator model builds multiple layers of latent variables as a prior model on top of the generator, which benefits learning complex data distribution and hierarchical representations. However, such a prior model usually focuses on modeling inter-layer relations between latent variables by assuming non-informative (conditional) Gaussian distributions, which can be limited in model expressivity. To tackle this issue and learn more expressive prior models, we propose an energy-based model (EBM) on the joint latent space over all layers of latent variables with the multi-layer generator as its backbone. Such joint latent space EBM prior model captures the intra-layer contextual relations at each layer through layer-wise energy terms, and latent variables across different layers are jointly corrected. We develop a joint training scheme via maximum likelihood estimation (MLE), which involves Markov Chain Monte Carlo (MCMC) sampling for both prior and posterior distributions of the latent variables from different layers. To ensure efficient inference and learning, we further propose a variational training scheme where an inference model is used to amortize the costly posterior MCMC sampling. Our experiments demonstrate that the learned model can be expressive in generating high-quality images …

It is commonly believed that more pre-training data leads to better transfer learning performance. However, recent evidence suggests that removing data from the source dataset can actually help too. In this work, we present a framework for probing the impact of the source dataset’s composition on transfer learning performance. Our framework facilitates new capabilities such as identifying transfer learning brittleness and detecting pathologies such as data-leakage and the presence of misleading examples in the source dataset. In particular, we demonstrate that removing detrimental datapoints identified by our framework improves transfer performance from ImageNet on a variety of transfer tasks.

We propose an approach to estimate the number of samples required for a model to reach a target performance. We find that the power law, the de facto principle to estimate model performance, leads to large error when using a small dataset (e.g., 5 samples per class) for extrapolation. This is because the log-performance error against the log-dataset size follows a nonlinear progression in the few-shot regime followed by a linear progression in the high-shot regime. We introduce a novel piecewise power law (PPL) that handles the two data regimes differently. To estimate the parameters of the PPL, we introduce a random forest regressor trained via meta learning that generalizes across classification/detection tasks, ResNet/ViT based architectures, and random/pre-trained initializations. The PPL improves the performance estimation on average by 37% across 16 classification datasets and 33% across 10 detection datasets, compared to the power law. We further extend the PPL to provide a confidence bound and use it to limit the prediction horizon that reduces over-estimation of data by 76% on classification and 91% on detection datasets.

Model selection is essential for reducing the search cost of the best pre-trained model over a large-scale model zoo for a downstream task. After analyzing recent hand-designed model selection criteria with 400+ ImageNet pre-trained models and 40 downstream tasks, we find that they can fail due to invalid assumptions and intrinsic limitations. The prior knowledge on model capacity and dataset also can not be easily integrated into the existing criteria. To address these issues, we propose to convert model selection as a recommendation problem and to learn from the past training history. Specifically, we characterize the meta information of datasets and models as features, and use their transfer learning performance as the guided score. With thousands of historical training jobs, a recommendation system can be learned to predict the model selection score given the features of the dataset and the model as input. Our approach enables integrating existing model selection scores as additional features and scales with more historical data. We evaluate the prediction accuracy with 22 pre-trained models over 40 downstream tasks. With extensive evaluations, we show that the learned approach can outperform prior hand-designed model selection methods significantly when relevant training history is available.

The success of multi-task learning (MTL) can largely be attributed to the shared representation of related tasks, allowing the models to better generalise. In deep learning, this is usually achieved by sharing a common neural network architecture and jointly training the weights. However, the joint training of weighting parameters on multiple related tasks may lead to performance degradation, known as negative transfer. To address this issue, this work proposes an evolutionary multi-tasking neural architecture search (EMT-NAS) algorithm to accelerate the search process by transferring architectural knowledge across multiple related tasks. In EMT-NAS, unlike the traditional MTL, the model for each task has a personalised network architecture and its own weights, thus offering the capability of effectively alleviating negative transfer. A fitness re-evaluation method is suggested to alleviate fluctuations in performance evaluations resulting from parameter sharing and the mini-batch gradient descent training method, thereby avoiding losing promising solutions during the search process. To rigorously verify the performance of EMT-NAS, the classification tasks used in the empirical assessments are derived from different datasets, including the CIFAR-10 and CIFAR-100, and four MedMNIST datasets. Extensive comparative experiments on different numbers of tasks demonstrate that EMT-NAS takes 8% and up to 40% on CIFAR and …

Continual learning aims to enable a model to incrementally learn knowledge from sequentially arrived data. Previous works adopt the conventional classification architecture, which consists of a feature extractor and a classifier. The feature extractor is shared across sequentially arrived tasks or classes, but one specific group of weights of the classifier corresponding to one new class should be incrementally expanded. Consequently, the parameters of a continual learner gradually increase. Moreover, as the classifier contains all historical arrived classes, a certain size of the memory is usually required to store rehearsal data to mitigate classifier bias and catastrophic forgetting. In this paper, we propose a non-incremental learner, named AttriCLIP, to incrementally extract knowledge of new classes or tasks. Specifically, AttriCLIP is built upon the pre-trained visual-language model CLIP. Its image encoder and text encoder are fixed to extract features from both images and text prompts. Each text prompt consists of a category name and a fixed number of learnable parameters which are selected from our designed attribute bank and serve as attributes. As we compute the visual and textual similarity for classification, AttriCLIP is a non-incremental learner. The attribute prompts, which encode the common knowledge useful for classification, can effectively mitigate …

In Continual Learning (CL), a model is required to learn a stream of tasks sequentially without significant performance degradation on previously learned tasks. Current approaches fail for a long sequence of tasks from diverse domains and difficulties. Many of the existing CL approaches are difficult to apply in practice due to excessive memory cost or training time, or are tightly coupled to a single device. With the intuition derived from the widely applied mini-batch training, we propose Batch Model Consolidation (BMC) to support more realistic CL under conditions where multiple agents are exposed to a range of tasks. During a regularization phase, BMC trains multiple expert models in parallel on a set of disjoint tasks. Each expert maintains weight similarity to a base model through a stability loss, and constructs a buffer from a fraction of the task’s data. During the consolidation phase, we combine the learned knowledge on ‘batches’ of expert models using a batched consolidation loss in memory data that aggregates all buffers. We thoroughly evaluate each component of our method in an ablation study and demonstrate the effectiveness on standardized benchmark datasets Split-CIFAR-100, Tiny-ImageNet, and the Stream dataset composed of 71 image classification tasks from diverse domains …

Existing methods mainly handle single weather types. However, the connections of different weather conditions at deep representation level are usually ignored. These connections, if used properly, can generate complementary representations for each other to make up insufficient training data, obtaining positive performance gains and better generalization. In this paper, we focus on the very correlated rain and snow to explore their connections at deep representation level. Because sub-optimal connections may cause negative effect, another issue is that if rain and snow are handled in a multi-task learning way, how to find an optimal connection strategy to simultaneously improve deraining and desnowing performance. To build desired connection, we propose a smart knowledge assignment strategy, called SmartAssign, to optimally assign the knowledge learned from both tasks to a specific one. In order to further enhance the accuracy of knowledge assignment, we propose a novel knowledge contrast mechanism, so that the knowledge assigned to different tasks preserves better uniqueness. The inherited inductive biases usually limit the modelling ability of CNNs, we introduce a novel transformer block to constitute the backbone of our network to effectively combine long-range context dependency and local image details. Extensive experiments on seven benchmark datasets verify that proposed SmartAssign …

Masked image modeling (MIM) performs strongly in pre-training large vision Transformers (ViTs). However, small models that are critical for real-world applications cannot or only marginally benefit from this pre-training approach. In this paper, we explore distillation techniques to transfer the success of large MIM-based pre-trained models to smaller ones. We systematically study different options in the distillation framework, including distilling targets, losses, input, network regularization, sequential distillation, etc, revealing that: 1) Distilling token relations is more effective than CLS token- and feature-based distillation; 2) An intermediate layer of the teacher network as target perform better than that using the last layer when the depth of the student mismatches that of the teacher; 3) Weak regularization is preferred; etc. With these findings, we achieve significant fine-tuning accuracy improvements over the scratch MIM pre-training on ImageNet-1K classification, using all the ViT-Tiny, ViT-Small, and ViT-base models, with +4.2%/+2.4%/+1.4% gains, respectively. Our TinyMIM model of base size achieves 52.2 mIoU in AE20K semantic segmentation, which is +4.1 higher than the MAE baseline. Our TinyMIM model of tiny size achieves 79.6% top-1 accuracy on ImageNet-1K image classification, which sets a new record for small vision models of the same size and computation budget. This strong …

Continual Learning (CL) aims to sequentially train models on streams of incoming data that vary in distribution by preserving previous knowledge while adapting to new data. Current CL literature focuses on restricted access to previously seen data, while imposing no constraints on the computational budget for training. This is unreasonable for applications in-the-wild, where systems are primarily constrained by computational and time budgets, not storage. We revisit this problem with a large-scale benchmark and analyze the performance of traditional CL approaches in a compute-constrained setting, where effective memory samples used in training can be implicitly restricted as a consequence of limited computation. We conduct experiments evaluating various CL sampling strategies, distillation losses, and partial fine-tuning on two large-scale datasets, namely ImageNet2K and Continual Google Landmarks V2 in data incremental, class incremental, and time incremental settings. Through extensive experiments amounting to a total of over 1500 GPU-hours, we find that, under compute-constrained setting, traditional CL approaches, with no exception, fail to outperform a simple minimal baseline that samples uniformly from memory. Our conclusions are consistent in a different number of stream time steps, e.g., 20 to 200, and under several computational budgets. This suggests that most existing CL methods are particularly …

Federated Learning (FL) has emerged as a de facto machine learning area and received rapid increasing research interests from the community. However, catastrophic forgetting caused by data heterogeneity and partial participation poses distinctive challenges for FL, which are detrimental to the performance. To tackle the problems, we propose a new FL approach (namely GradMA), which takes inspiration from continual learning to simultaneously correct the server-side and worker-side update directions as well as take full advantage of server’s rich computing and memory resources. Furthermore, we elaborate a memory reduction strategy to enable GradMA to accommodate FL with a large scale of workers. We then analyze convergence of GradMA theoretically under the smooth non-convex setting and show that its convergence rate achieves a linear speed up w.r.t the increasing number of sampled active workers. At last, our extensive experiments on various image classification tasks show that GradMA achieves significant performance gains in accuracy and communication efficiency compared to SOTA baselines. We provide our code here: https://github.com/lkyddd/GradMA.

Continual learning aims to incrementally learn novel classes over time, while not forgetting the learned knowledge. Recent studies have found that learning would not forget if the updated gradient is orthogonal to the feature space. However, previous approaches require the gradient to be fully orthogonal to the whole feature space, leading to poor plasticity, as the feasible gradient direction becomes narrow when the tasks continually come, i.e., feature space is unlimitedly expanded. In this paper, we propose a space decoupling (SD) algorithm to decouple the feature space into a pair of complementary subspaces, i.e., the stability space I, and the plasticity space R. I is established by conducting space intersection between the historic and current feature space, and thus I contains more task-shared bases. R is constructed by seeking the orthogonal complementary subspace of I, and thus R mainly contains more task-specific bases. By putting the distinguishing constraints on R and I, our method achieves a better balance between stability and plasticity. Extensive experiments are conducted by applying SD to gradient projection baselines, and show SD is model-agnostic and achieves SOTA results on publicly available datasets.

Fully test-time adaptation aims to adapt the network model based on sequential analysis of input samples during the inference stage to address the cross-domain performance degradation problem of deep neural networks. We take inspiration from the biological plausibility learning where the neuron responses are tuned based on a local synapse-change procedure and activated by competitive lateral inhibition rules. Based on these feed-forward learning rules, we design a soft Hebbian learning process which provides an unsupervised and effective mechanism for online adaptation. We observe that the performance of this feed-forward Hebbian learning for fully test-time adaptation can be significantly improved by incorporating a feedback neuro-modulation layer. It is able to fine-tune the neuron responses based on the external feedback generated by the error back-propagation from the top inference layers. This leads to our proposed neuro-modulated Hebbian learning (NHL) method for fully test-time adaptation. With the unsupervised feed-forward soft Hebbian learning being combined with a learned neuro-modulator to capture feedback from external responses, the source model can be effectively adapted during the testing process. Experimental results on benchmark datasets demonstrate that our proposed method can significantly improve the adaptation performance of network models and outperforms existing state-of-the-art methods.

Dataset Distillation aims to distill an entire dataset’s knowledge into a few synthetic images. The idea is to synthesize a small number of synthetic data points that, when given to a learning algorithm as training data, result in a model approximating one trained on the original data. Despite a recent upsurge of progress in the field, existing dataset distillation methods fail to generalize to new architectures and scale to high-resolution datasets. To overcome the above issues, we propose to use the learned prior from pre-trained deep generative models to synthesize the distilled data. To achieve this, we present a new optimization algorithm that distills a large number of images into a few intermediate feature vectors in the generative model’s latent space. Our method augments existing techniques, significantly improving cross-architecture generalization in all settings.

Model-based deep learning has achieved astounding successes due in part to the availability of large-scale real-world data. However, processing such massive amounts of data comes at a considerable cost in terms of computations, storage, training and the search for good neural architectures. Dataset distillation has thus recently come to the fore. This paradigm involves distilling information from large real-world datasets into tiny and compact synthetic datasets such that processing the latter yields similar performances as the former. State-of-the-art methods primarily rely on learning the synthetic dataset by matching the gradients obtained during training between the real and synthetic data. However, these gradient-matching methods suffer from the accumulated trajectory error caused by the discrepancy between the distillation and subsequent evaluation. To alleviate the adverse impact of this accumulated trajectory error, we propose a novel approach that encourages the optimization algorithm to seek a flat trajectory. We show that the weights trained on synthetic data are robust against the accumulated errors perturbations with the regularization towards the flat trajectory. Our method, called Flat Trajectory Distillation (FTD), is shown to boost the performance of gradient-matching methods by up to 4.7% on a subset of images of the ImageNet dataset with higher resolution images. …

Dataset distillation, also known as dataset condensation, aims to compress a large dataset into a compact synthetic one. Existing methods perform dataset condensation by assuming a fixed storage or transmission budget. When the budget changes, however, they have to repeat the synthesizing process with access to original datasets, which is highly cumbersome if not infeasible at all. In this paper, we explore the problem of slimmable dataset condensation, to extract a smaller synthetic dataset given only previous condensation results. We first study the limitations of existing dataset condensation algorithms on such a successive compression setting and identify two key factors: (1) the inconsistency of neural networks over different compression times and (2) the underdetermined solution space for synthetic data. Accordingly, we propose a novel training objective for slimmable dataset condensation to explicitly account for both factors. Moreover, synthetic datasets in our method adopt an significance-aware parameterization. Theoretical derivation indicates that an upper-bounded error can be achieved by discarding the minor components without training. Alternatively, if training is allowed, this strategy can serve as a strong initialization that enables a fast convergence. Extensive comparisons and ablations demonstrate the superiority of the proposed solution over existing methods on multiple benchmarks.

The goal of domain generalization (DG) is to enhance the generalization capability of the model learned from a source domain to other unseen domains. The recently developed Sharpness-Aware Minimization (SAM) method aims to achieve this goal by minimizing the sharpness measure of the loss landscape. Though SAM and its variants have demonstrated impressive DG performance, they may not always converge to the desired flat region with a small loss value. In this paper, we present two conditions to ensure that the model could converge to a flat minimum with a small loss, and present an algorithm, named Sharpness-Aware Gradient Matching (SAGM), to meet the two conditions for improving model generalization capability. Specifically, the optimization objective of SAGM will simultaneously minimize the empirical risk, the perturbed loss (i.e., the maximum loss within a neighborhood in the parameter space), and the gap between them. By implicitly aligning the gradient directions between the empirical risk and the perturbed loss, SAGM improves the generalization capability over SAM and its variants without increasing the computational cost. Extensive experimental results show that our proposed SAGM method consistently outperforms the state-of-the-art methods on five DG benchmarks, including PACS, VLCS, OfficeHome, TerraIncognita, and DomainNet. Codes are available at …

In this paper, we present a new MTL framework that searches for structures optimized for multiple tasks with diverse graph topologies and shares features among tasks. We design a restricted DAG-based central network with read-in/read-out layers to build topologically diverse task-adaptive structures while limiting search space and time. We search for a single optimized network that serves as multiple task adaptive sub-networks using our three-stage training process. To make the network compact and discretized, we propose a flow-based reduction algorithm and a squeeze loss used in the training process. We evaluate our optimized network on various public MTL datasets and show ours achieves state-of-the-art performance. An extensive ablation study experimentally validates the effectiveness of the sub-module and schemes in our framework.

Current Deep Network (DN) visualization and interpretability methods rely heavily on data space visualizations such as scoring which dimensions of the data are responsible for their associated prediction or generating new data features or samples that best match a given DN unit or representation. In this paper, we go one step further by developing the first provably exact method for computing the geometry of a DN’s mapping -- including its decision boundary -- over a specified region of the data space. By leveraging the theory of Continuous Piecewise Linear (CPWL) spline DNs, SplineCam exactly computes a DN’s geometry without resorting to approximations such as sampling or architecture simplification. SplineCam applies to any DN architecture based on CPWL activation nonlinearities, including (leaky) ReLU, absolute value, maxout, and max-pooling and can also be applied to regression DNs such as implicit neural representations. Beyond decision boundary visualization and characterization, SplineCam enables one to compare architectures, measure generalizability, and sample from the decision boundary on or off the data manifold. Project website: https://bit.ly/splinecam

Since the introduction of deep learning, a wide scope of representation properties, such as decorrelation, whitening, disentanglement, rank, isotropy, and mutual information, have been studied to improve the quality of representation. However, manipulating such properties can be challenging in terms of implementational effectiveness and general applicability. To address these limitations, we propose to regularize von Neumann entropy (VNE) of representation. First, we demonstrate that the mathematical formulation of VNE is superior in effectively manipulating the eigenvalues of the representation autocorrelation matrix. Then, we demonstrate that it is widely applicable in improving state-of-the-art algorithms or popular benchmark algorithms by investigating domain-generalization, meta-learning, self-supervised learning, and generative models. In addition, we formally establish theoretical connections with rank, disentanglement, and isotropy of representation. Finally, we provide discussions on the dimension control of VNE and the relationship with Shannon entropy. Code is available at: https://github.com/jaeill/CVPR23-VNE.

Despite its importance for federated learning, continuous learning and many other applications, on-device training remains an open problem for EdgeAI. The problem stems from the large number of operations (e.g., floating point multiplications and additions) and memory consumption required during training by the back-propagation algorithm. Consequently, in this paper, we propose a new gradient filtering approach which enables on-device CNN model training. More precisely, our approach creates a special structure with fewer unique elements in the gradient map, thus significantly reducing the computational complexity and memory consumption of back propagation during training. Extensive experiments on image classification and semantic segmentation with multiple CNN models (e.g., MobileNet, DeepLabV3, UPerNet) and devices (e.g., Raspberry Pi and Jetson Nano) demonstrate the effectiveness and wide applicability of our approach. For example, compared to SOTA, we achieve up to 19× speedup and 77.1% memory savings on ImageNet classification with only 0.1% accuracy loss. Finally, our method is easy to implement and deploy; over 20× speedup and 90% energy savings have been observed compared to highly optimized baselines in MKLDNN and CUDNN on NVIDIA Jetson Nano. Consequently, our approach opens up a new direction of research with a huge potential for on-device training.

As black-box models increasingly power high-stakes applications, a variety of data-driven explanation methods have been introduced. Meanwhile, machine learning models are constantly challenged by distributional shifts. A question naturally arises: Are data-driven explanations robust against out-of-distribution data? Our empirical results show that even though predict correctly, the model might still yield unreliable explanations under distributional shifts. How to develop robust explanations against out-of-distribution data? To address this problem, we propose an end-to-end model-agnostic learning framework Distributionally Robust Explanations (DRE). The key idea is, inspired by self-supervised learning, to fully utilizes the inter-distribution information to provide supervisory signals for the learning of explanations without human annotation. Can robust explanations benefit the model’s generalization capability? We conduct extensive experiments on a wide range of tasks and data types, including classification and regression on image and scientific tabular data. Our results demonstrate that the proposed method significantly improves the model’s performance in terms of explanation and prediction robustness against distributional shifts.

Neural networks are often prone to bias toward spurious correlations inherent in a dataset, thus failing to generalize unbiased test criteria. A key challenge to resolving the issue is the significant lack of bias-conflicting training data (i.e., samples without spurious correlations). In this paper, we propose a novel data augmentation approach termed Bias-Adversarial augmentation (BiasAdv) that supplements bias-conflicting samples with adversarial images. Our key idea is that an adversarial attack on a biased model that makes decisions based on spurious correlations may generate synthetic bias-conflicting samples, which can then be used as augmented training data for learning a debiased model. Specifically, we formulate an optimization problem for generating adversarial images that attack the predictions of an auxiliary biased model without ruining the predictions of the desired debiased model. Despite its simplicity, we find that BiasAdv can generate surprisingly useful synthetic bias-conflicting samples, allowing the debiased model to learn generalizable representations. Furthermore, BiasAdv does not require any bias annotations or prior knowledge of the bias type, which enables its broad applicability to existing debiasing methods to improve their performances. Our extensive experimental results demonstrate the superiority of BiasAdv, achieving state-of-the-art performance on four popular benchmark datasets across various bias domains.

The recent detection transformer (DETR) has advanced object detection, but its application on resource-constrained devices requires massive computation and memory resources. Quantization stands out as a solution by representing the network in low-bit parameters and operations. However, there is a significant performance drop when performing low-bit quantized DETR (Q-DETR) with existing quantization methods. We find that the bottlenecks of Q-DETR come from the query information distortion through our empirical analyses. This paper addresses this problem based on a distribution rectification distillation (DRD). We formulate our DRD as a bi-level optimization problem, which can be derived by generalizing the information bottleneck (IB) principle to the learning of Q-DETR. At the inner level, we conduct a distribution alignment for the queries to maximize the self-information entropy. At the upper level, we introduce a new foreground-aware query matching scheme to effectively transfer the teacher information to distillation-desired features to minimize the conditional information entropy. Extensive experimental results show that our method performs much better than prior arts. For example, the 4-bit Q-DETR can theoretically accelerate DETR with ResNet-50 backbone by 6.6x and achieve 39.4% AP, with only 2.6% performance gaps than its real-valued counterpart on the COCO dataset.

Straight-through estimator (STE), which enables the gradient flow over the non-differentiable function via approximation, has been favored in studies related to quantization-aware training (QAT). However, STE incurs unstable convergence during QAT, resulting in notable quality degradation in low-precision representation. Recently, pseudo-quantization training has been proposed as an alternative approach to updating the learnable parameters using the pseudo-quantization noise instead of STE. In this study, we propose a novel noise proxy-based integrated pseudo-quantization (NIPQ) that enables unified support of pseudo-quantization for both activation and weight with minimal error by integrating the idea of truncation on the pseudo-quantization framework. NIPQ updates all of the quantization parameters (e.g., bit-width and truncation boundary) as well as the network parameters via gradient descent without STE instability, resulting in greatly-simplified but reliable precision allocation without human intervention. Our extensive experiments show that NIPQ outperforms existing quantization algorithms in various vision and language applications by a large margin.

Large-scale training of modern deep learning models heavily relies on publicly available data on the web. This potentially unauthorized usage of online data leads to concerns regarding data privacy. Recent works aim to make unlearnable data for deep learning models by adding small, specially designed noises to tackle this issue. However, these methods are vulnerable to adversarial training (AT) and/or are computationally heavy. In this work, we propose a novel, model-free, Convolution-based Unlearnable DAtaset (CUDA) generation technique. CUDA is generated using controlled class-wise convolutions with filters that are randomly generated via a private key. CUDA encourages the network to learn the relation between filters and labels rather than informative features for classifying the clean data. We develop some theoretical analysis demonstrating that CUDA can successfully poison Gaussian mixture data by reducing the clean data performance of the optimal Bayes classifier. We also empirically demonstrate the effectiveness of CUDA with various datasets (CIFAR-10, CIFAR-100, ImageNet-100, and Tiny-ImageNet), and architectures (ResNet-18, VGG-16, Wide ResNet-34-10, DenseNet-121, DeIT, EfficientNetV2-S, and MobileNetV2). Our experiments show that CUDA is robust to various data augmentations and training approaches such as smoothing, AT with different budgets, transfer learning, and fine-tuning. For instance, training a ResNet-18 on ImageNet-100 CUDA …

Generative Adversarial Networks (GANs) rely heavily on large-scale training data for training high-quality image generation models. With limited training data, the GAN discriminator often suffers from severe overfitting which directly leads to degraded generation especially in generation diversity. Inspired by the recent advances in knowledge distillation (KD), we propose KD-GAN, a knowledge-distillation based generation framework that introduces pre-trained vision-language models for training effective data-limited image generation models. KD-GAN consists of two innovative designs. The first is aggregated generative KD that mitigates the discriminator overfitting by challenging the discriminator with harder learning tasks and distilling more generalizable knowledge from the pre-trained models. The second is correlated generative KD that improves the generation diversity by distilling and preserving the diverse image-text correlation within the pre-trained models. Extensive experiments over multiple benchmarks show that KD-GAN achieves superior image generation with limited training data. In addition, KD-GAN complements the state-of-the-art with consistent and substantial performance gains. Note that codes will be released.

Training Generative adversarial networks (GANs) stably is a challenging task. The generator in GANs transform noise vectors, typically Gaussian distributed, into realistic data such as images. In this paper, we propose a novel approach for training GANs with images as inputs, but without enforcing any pairwise constraints. The intuition is that images are more structured than noise, which the generator can leverage to learn a more robust transformation. The process can be made efficient by identifying closely related datasets, or a “friendly neighborhood” of the target distribution, inspiring the moniker, Spider GAN. To define friendly neighborhoods leveraging proximity between datasets, we propose a new measure called the signed inception distance (SID), inspired by the polyharmonic kernel. We show that the Spider GAN formulation results in faster convergence, as the generator can discover correspondence even between seemingly unrelated datasets, for instance, between Tiny-ImageNet and CelebA faces. Further, we demonstrate cascading Spider GAN, where the output distribution from a pre-trained GAN generator is used as the input to the subsequent network. Effectively, transporting one distribution to another in a cascaded fashion until the target is learnt -- a new flavor of transfer learning. We demonstrate the efficacy of the Spider approach on …

The deployment of perception systems based on neural networks in safety critical applications requires assurance on their robustness. Deterministic guarantees on network robustness require formal verification. Standard approaches for verifying robustness analyse invariance to analytically defined transformations, but not the diverse and ubiquitous changes involving object pose, scene viewpoint, occlusions, etc. To this end, we present an efficient approach for verifying specifications definable using Latent Variable Models that capture such diverse changes. The approach involves adding an invertible encoding head to the network to be verified, enabling the verification of latent space sets with minimal reconstruction overhead. We report verification experiments for three classes of proposed latent space specifications, each capturing different types of realistic input variations. Differently from previous work in this area, the proposed approach is relatively independent of input dimensionality and scales to a broad class of deep networks and real-world datasets by mitigating the inefficiency and decoder expressivity dependence in the present state-of-the-art.

The cost of pathological examination makes virtual re-staining of pathological images meaningful. However, due to the ultra-high resolution of pathological images, traditional virtual re-staining methods have to divide a WSI image into patches for model training and inference. Such a limitation leads to the lack of global information, resulting in observable differences in color, brightness and contrast when the re-stained patches are merged to generate an image of larger size. We summarize this issue as the square effect. Some existing methods try to solve this issue through overlapping between patches or simple post-processing. But the former one is not that effective, while the latter one requires carefully tuning. In order to eliminate the square effect, we design a bi-directional feature fusion generative adversarial network (BFF-GAN) with a global branch and a local branch. It learns the inter-patch connections through the fusion of global and local features plus patch-wise attention. We perform experiments on both the private dataset RCC and the public dataset ANHIR. The results show that our model achieves competitive performance and is able to generate extremely real images that are deceptive even for experienced pathologists, which means it is of great clinical significance.

Visual anomaly detection, an important problem in computer vision, is usually formulated as a one-class classification and segmentation task. The student-teacher (S-T) framework has proved to be effective in solving this challenge. However, previous works based on S-T only empirically applied constraints on normal data and fused multi-level information. In this study, we propose an improved model called DeSTSeg, which integrates a pre-trained teacher network, a denoising student encoder-decoder, and a segmentation network into one framework. First, to strengthen the constraints on anomalous data, we introduce a denoising procedure that allows the student network to learn more robust representations. From synthetically corrupted normal images, we train the student network to match the teacher network feature of the same images without corruption. Second, to fuse the multi-level S-T features adaptively, we train a segmentation network with rich supervision from synthetic anomaly masks, achieving a substantial performance improvement. Experiments on the industrial inspection benchmark dataset demonstrate that our method achieves state-of-the-art performance, 98.6% on image-level AUC, 75.8% on pixel-level average precision, and 76.4% on instance-level average precision.

Unsupervised anomaly localization and detection is crucial for industrial manufacturing processes due to the lack of anomalous samples. Recent unsupervised advances on industrial anomaly detection achieve high performance by training separate models for many different categories. The model storage and training time cost of this paradigm is high. Moreover, the setting of one-model-N-classes leads to fearful degradation of existing methods. In this paper, we propose a unified CNN framework for unsupervised anomaly localization, named OmniAL. This method conquers aforementioned problems by improving anomaly synthesis, reconstruction and localization. To prevent the model learning identical reconstruction, it trains the model with proposed panel-guided synthetic anomaly data rather than directly using normal data. It increases anomaly reconstruction error for multi-class distribution by using a network that is equipped with proposed Dilated Channel and Spatial Attention (DCSA) blocks. To better localize the anomaly regions, it employs proposed DiffNeck between reconstruction and localization sub-networks to explore multi-level differences. Experiments on 15-class MVTecAD and 12-class VisA datasets verify the advantage of proposed OmniAL that surpasses the state-of-the-art of unified models. On 15-class-MVTecAD/12-class-VisA, its single unified model achieves 97.2/87.8 image-AUROC, 98.3/96.6 pixel-AUROC and 73.4/41.7 pixel-AP for anomaly detection and localization respectively. Besides that, we make the first …

Federated learning-based semantic segmentation (FSS) has drawn widespread attention via decentralized training on local clients. However, most FSS models assume categories are fxed in advance, thus heavily undergoing forgetting on old categories in practical applications where local clients receive new categories incrementally while have no memory storage to access old classes. Moreover, new clients collecting novel classes may join in the global training of FSS, which further exacerbates catastrophic forgetting. To surmount the above challenges, we propose a Forgetting-Balanced Learning (FBL) model to address heterogeneous forgetting on old classes from both intra-client and inter-client aspects. Specifically, under the guidance of pseudo labels generated via adaptive class-balanced pseudo labeling, we develop a forgetting-balanced semantic compensation loss and a forgetting-balanced relation consistency loss to rectify intra-client heterogeneous forgetting of old categories with background shift. It performs balanced gradient propagation and relation consistency distillation within local clients. Moreover, to tackle heterogeneous forgetting from inter-client aspect, we propose a task transition monitor. It can identify new classes under privacy protection and store the latest old global model for relation distillation. Qualitative experiments reveal large improvement of our model against comparison methods. The code is available at https://github.com/JiahuaDong/FISS.

Although federated learning has made awe-inspiring advances, most studies have assumed that the client’s data are fully labeled. However, in a real-world scenario, every client may have a significant amount of unlabeled instances. Among the various approaches to utilizing unlabeled data, a federated active learning framework has emerged as a promising solution. In the decentralized setting, there are two types of available query selector models, namely ‘global’ and ‘local-only’ models, but little literature discusses their performance dominance and its causes. In this work, we first demonstrate that the superiority of two selector models depends on the global and local inter-class diversity. Furthermore, we observe that the global and local-only models are the keys to resolving the imbalance of each side. Based on our findings, we propose LoGo, a FAL sampling strategy robust to varying local heterogeneity levels and global imbalance ratio, that integrates both models by two steps of active selection scheme. LoGo consistently outperforms six active learning strategies in the total number of 38 experimental settings.

Federated Domain Generalization (FedDG) attempts to learn a global model in a privacy-preserving manner that generalizes well to new clients possibly with domain shift. Recent exploration mainly focuses on designing an unbiased training strategy within each individual domain. However, without the support of multi-domain data jointly in the mini-batch training, almost all methods cannot guarantee the generalization under domain shift. To overcome this problem, we propose a novel global objective incorporating a new variance reduction regularizer to encourage fairness. A novel FL-friendly method named Generalization Adjustment (GA) is proposed to optimize the above objective by dynamically calibrating the aggregation weights. The theoretical analysis of GA demonstrates the possibility to achieve a tighter generalization bound with an explicit re-weighted aggregation, substituting the implicit multi-domain data sharing that is only applicable to the conventional DG settings. Besides, the proposed algorithm is generic and can be combined with any local client training-based methods. Extensive experiments on several benchmark datasets have shown the effectiveness of the proposed method, with consistent improvements over several FedDG algorithms when used in combination. The source code is released at https://github.com/MediaBrain-SJTU/FedDG-GA.

Data heterogeneity across clients is a key challenge in federated learning. Prior works address this by either aligning client and server models or using control variates to correct client model drift. Although these methods achieve fast convergence in convex or simple non-convex problems, the performance in over-parameterized models such as deep neural networks is lacking. In this paper, we first revisit the widely used FedAvg algorithm in a deep neural network to understand how data heterogeneity influences the gradient updates across the neural network layers. We observe that while the feature extraction layers are learned efficiently by FedAvg, the substantial diversity of the final classification layers across clients impedes the performance. Motivated by this, we propose to correct model drift by variance reduction only on the final layers. We demonstrate that this significantly outperforms existing benchmarks at a similar or lower communication cost. We furthermore provide proof for the convergence rate of our algorithm.

Secure aggregation promises a heightened level of privacy in federated learning, maintaining that a server only has access to a decrypted aggregate update. Within this setting, linear layer leakage methods are the only data reconstruction attacks able to scale and achieve a high leakage rate regardless of the number of clients or batch size. This is done through increasing the size of an injected fully-connected (FC) layer. We show that this results in a resource overhead which grows larger with an increasing number of clients. We show that this resource overhead is caused by an incorrect perspective in all prior work that treats an attack on an aggregate update in the same way as an individual update with a larger batch size. Instead, by attacking the update from the perspective that aggregation is combining multiple individual updates, this allows the application of sparsity to alleviate resource overhead. We show that the use of sparsity can decrease the model size overhead by over 327× and the computation time by 3.34× compared to SOTA while maintaining equivalent total leakage rate, 77% even with 1000 clients in aggregation.

There is a growing interest in developing unlearnable examples (UEs) against visual privacy leaks on the Internet. UEs are training samples added with invisible but unlearnable noise, which have been found can prevent unauthorized training of machine learning models. UEs typically are generated via a bilevel optimization framework with a surrogate model to remove (minimize) errors from the original samples, and then applied to protect the data against unknown target models. However, existing UE generation methods all rely on an ideal assumption called labelconsistency, where the hackers and protectors are assumed to hold the same label for a given sample. In this work, we propose and promote a more practical label-agnostic setting, where the hackers may exploit the protected data quite differently from the protectors. E.g., a m-class unlearnable dataset held by the protector may be exploited by the hacker as a n-class dataset. Existing UE generation methods are rendered ineffective in this challenging setting. To tackle this challenge, we present a novel technique called Unlearnable Clusters (UCs) to generate label-agnostic unlearnable examples with cluster-wise perturbations. Furthermore, we propose to leverage Vision-and-Language Pretrained Models (VLPMs) like CLIP as the surrogate model to improve the transferability of the crafted UCs to …

In this paper, we analyse the generalization ability of binary classifiers for the task of deepfake detection. We find that the stumbling block to their generalization is caused by the unexpected learned identity representation on images. Termed as the Implicit Identity Leakage, this phenomenon has been qualitatively and quantitatively verified among various DNNs. Furthermore, based on such understanding, we propose a simple yet effective method named the ID-unaware Deepfake Detection Model to reduce the influence of this phenomenon. Extensive experimental results demonstrate that our method outperforms the state-of-the-art in both in-dataset and cross-dataset evaluation. The code is available at https://github.com/megvii-research/CADDM.

Backdoor defenses have been studied to alleviate the threat of deep neural networks (DNNs) being backdoor attacked and thus maliciously altered. Since DNNs usually adopt some external training data from an untrusted third party, a robust backdoor defense strategy during the training stage is of importance. We argue that the core of training-time defense is to select poisoned samples and to handle them properly. In this work, we summarize the training-time defenses from a unified framework as splitting the poisoned dataset into two data pools. Under our framework, we propose an adaptively splitting dataset-based defense (ASD). Concretely, we apply loss-guided split and meta-learning-inspired split to dynamically update two data pools. With the split clean data pool and polluted data pool, ASD successfully defends against backdoor attacks during training. Extensive experiments on multiple benchmark datasets and DNN models against six state-of-the-art backdoor attacks demonstrate the superiority of our ASD.

Diffusion models are state-of-the-art deep learning empowered generative models that are trained based on the principle of learning forward and reverse diffusion processes via progressive noise-addition and denoising. To gain a better understanding of the limitations and potential risks, this paper presents the first study on the robustness of diffusion models against backdoor attacks. Specifically, we propose BadDiffusion, a novel attack framework that engineers compromised diffusion processes during model training for backdoor implantation. At the inference stage, the backdoored diffusion model will behave just like an untampered generator for regular data inputs, while falsely generating some targeted outcome designed by the bad actor upon receiving the implanted trigger signal. Such a critical risk can be dreadful for downstream tasks and applications built upon the problematic model. Our extensive experiments on various backdoor attack settings show that BadDiffusion can consistently lead to compromised diffusion models with high utility and target specificity. Even worse, BadDiffusion can be made cost-effective by simply finetuning a clean pre-trained diffusion model to implant backdoors. We also explore some possible countermeasures for risk mitigation. Our results call attention to potential risks and possible misuse of diffusion models.

Vision Transformers (ViTs) have demonstrated the state-of-the-art performance in various vision-related tasks. The success of ViTs motivates adversaries to perform backdoor attacks on ViTs. Although the vulnerability of traditional CNNs to backdoor attacks is well-known, backdoor attacks on ViTs are seldom-studied. Compared to CNNs capturing pixel-wise local features by convolutions, ViTs extract global context information through patches and attentions. Naïvely transplanting CNN-specific backdoor attacks to ViTs yields only a low clean data accuracy and a low attack success rate. In this paper, we propose a stealth and practical ViT-specific backdoor attack TrojViT. Rather than an area-wise trigger used by CNN-specific backdoor attacks, TrojViT generates a patch-wise trigger designed to build a Trojan composed of some vulnerable bits on the parameters of a ViT stored in DRAM memory through patch salience ranking and attention-target loss. TrojViT further uses parameter distillation to reduce the bit number of the Trojan. Once the attacker inserts the Trojan into the ViT model by flipping the vulnerable bits, the ViT model still produces normal inference accuracy with benign inputs. But when the attacker embeds a trigger into an input, the ViT model is forced to classify the input to a predefined target class. We show that …

Diffusion models have achieved great success in a range of tasks, such as image synthesis and molecule design. As such successes hinge on large-scale training data collected from diverse sources, the trustworthiness of these collected data is hard to control or audit. In this work, we aim to explore the vulnerabilities of diffusion models under potential training data manipulations and try to answer: How hard is it to perform Trojan attacks on well-trained diffusion models? What are the adversarial targets that such Trojan attacks can achieve? To answer these questions, we propose an effective Trojan attack against diffusion models, TrojDiff, which optimizes the Trojan diffusion and generative processes during training. In particular, we design novel transitions during the Trojan diffusion process to diffuse adversarial targets into a biased Gaussian distribution and propose a new parameterization of the Trojan generative process that leads to an effective training objective for the attack. In addition, we consider three types of adversarial targets: the Trojaned diffusion models will always output instances belonging to a certain class from the in-domain distribution (In-D2D attack), out-of-domain distribution (Out-D2D-attack), and one specific instance (D2I attack). We evaluate TrojDiff on CIFAR-10 and CelebA datasets against both DDPM and DDIM …

We propose an approach for adversarial attacks on dense prediction models (such as object detectors and segmentation). It is well known that the attacks generated by a single surrogate model do not transfer to arbitrary (blackbox) victim models. Furthermore, targeted attacks are often more challenging than the untargeted attacks. In this paper, we show that a carefully designed ensemble can create effective attacks for a number of victim models. In particular, we show that normalization of the weights for individual models plays a critical role in the success of the attacks. We then demonstrate that by adjusting the weights of the ensemble according to the victim model can further improve the performance of the attacks. We performed a number of experiments for object detectors and segmentation to highlight the significance of the our proposed methods. Our proposed ensemble-based method outperforms existing blackbox attack methods for object detection and segmentation. Finally we show that our proposed method can also generate a single perturbation that can fool multiple blackbox detection and segmentation models simultaneously.

Attackers can deceive neural networks by adding human imperceptive perturbations to their input data; this reveals the vulnerability and weak robustness of current deep-learning networks. Many attack techniques have been proposed to evaluate the model’s robustness. Gradient-based attacks suffer from severely overestimating the robustness. This paper identifies that the relative error in calculated gradients caused by floating-point errors, including floating-point underflow and rounding errors, is a fundamental reason why gradient-based attacks fail to accurately assess the model’s robustness. Although it is hard to eliminate the relative error in the gradients, we can control its effect on the gradient-based attacks. Correspondingly, we propose an efficient loss function by minimizing the detrimental impact of the floating-point errors on the attacks. Experimental results show that it is more efficient and reliable than other loss functions when examined across a wide range of defence mechanisms.

Many defenses against adversarial attacks (e.g. robust classifiers, randomization, or image purification) use countermeasures put to work only after the attack has been crafted. We adopt a different perspective to introduce A^5 (Adversarial Augmentation Against Adversarial Attacks), a novel framework including the first certified preemptive defense against adversarial attacks. The main idea is to craft a defensive perturbation to guarantee that any attack (up to a given magnitude) towards the input in hand will fail. To this aim, we leverage existing automatic perturbation analysis tools for neural networks. We study the conditions to apply A^5 effectively, analyze the importance of the robustness of the to-be-defended classifier, and inspect the appearance of the robustified images. We show effective on-the-fly defensive augmentation with a robustifier network that ignores the ground truth label, and demonstrate the benefits of robustifier and classifier co-training. In our tests, A^5 consistently beats state of the art certified defenses on MNIST, CIFAR10, FashionMNIST and Tinyimagenet. We also show how to apply A^5 to create certifiably robust physical objects. The released code at https://github.com/NVlabs/A5 allows experimenting on a wide range of scenarios beyond the man-in-the-middle attack tested here, including the case of physical attacks.

Deep Neural Networks show superior performance in various tasks but are vulnerable to adversarial attacks. Most defense techniques are devoted to the adversarial training strategies, however, it is difficult to achieve satisfactory robust performance only with traditional adversarial training. We mainly attribute it to that aggressive perturbations which lead to the loss increment can always be found via gradient ascent in white-box setting. Although some noises can be involved to prevent attacks from deriving precise gradients on inputs, there exist trade-offs between the defense capability and natural generalization. Taking advantage of the properties of random projection, we propose to replace part of convolutional filters with random projection filters, and theoretically explore the geometric representation preservation of proposed synthesized filters via Johnson-Lindenstrauss lemma. We conduct sufficient evaluation on multiple networks and datasets. The experimental results showcase the superiority of proposed random projection filters to state-of-the-art baselines. The code is available on https://github.com/UniSerj/Random-Projection-Filters.

Real-world adversarial physical patches were recently shown to be successful in compromising state-of-the-art models in a variety of computer vision applications. The most promising defenses that are based on either input gradient or features analyses have been shown to be compromised by recent GAN-based adaptive attacks that generate realistic/naturalistic patches. In this paper, we propose Jedi, a new defense against adversarial patches that is resilient to realistic patch attacks, and also improves detection and recovery compared to the state of the art. Jedi leverages two new ideas: (1) it improves the identification of potential patch regions using entropy analysis: we show that the entropy of adversarial patches is high, even in naturalistic patches; and (2) it improves the localization of adversarial patches, using an autoencoder that is able to complete patch regions and filter out normal regions with high entropy that are not part of a patch. Jedi achieves high precision adversarial patch localization, which we show is critical to successfully repair the images. Since Jedi relies on an input entropy analysis, it is model-agnostic, and can be applied on pre-trained off-the-shelf models without changes to the training or inference of the protected models. Jedi detects on average 90% of …

Adversarial training has been demonstrated to be one of the most effective remedies for defending adversarial examples, yet it often suffers from the huge robustness generalization gap on unseen testing adversaries, deemed as the adversarially robust generalization problem. Despite the preliminary understandings devoted to adversarially robust generalization, little is known from the architectural perspective. To bridge the gap, this paper for the first time systematically investigated the relationship between adversarially robust generalization and architectural design. In particular, we comprehensively evaluated 20 most representative adversarially trained architectures on ImageNette and CIFAR-10 datasets towards multiple l_p-norm adversarial attacks. Based on the extensive experiments, we found that, under aligned settings, Vision Transformers (e.g., PVT, CoAtNet) often yield better adversarially robust generalization while CNNs tend to overfit on specific attacks and fail to generalize on multiple adversaries. To better understand the nature behind it, we conduct theoretical analysis via the lens of Rademacher complexity. We revealed the fact that the higher weight sparsity contributes significantly towards the better adversarially robust generalization of Transformers, which can be often achieved by the specially-designed attention blocks. We hope our paper could help to better understand the mechanism for designing robust DNNs. Our model weights can be found …

Despite their success, vision transformers still remain vulnerable to image corruptions, such as noise or blur. Indeed, we find that the vulnerability mainly stems from the unstable self-attention mechanism, which is inherently built upon patch-based inputs and often becomes overly sensitive to the corruptions across patches. For example, when we only occlude a small number of patches with random noise (e.g., 10%), these patch corruptions would lead to severe accuracy drops and greatly distract intermediate attention layers. To address this, we propose a new training method that improves the robustness of transformers from a new perspective -- reducing sensitivity to patch corruptions (RSPC). Specifically, we first identify and occlude/corrupt the most vulnerable patches and then explicitly reduce sensitivity to them by aligning the intermediate features between clean and corrupted examples. We highlight that the construction of patch corruptions is learned adversarially to the following feature alignment process, which is particularly effective and essentially different from existing methods. In experiments, our RSPC greatly improves the stability of attention layers and consistently yields better robustness on various benchmarks, including CIFAR-10/100-C, ImageNet-A, ImageNet-C, and ImageNet-P.

Face recognition is a prevailing authentication solution in numerous biometric applications. Physical adversarial attacks, as an important surrogate, can identify the weaknesses of face recognition systems and evaluate their robustness before deployed. However, most existing physical attacks are either detectable readily or ineffective against commercial recognition systems. The goal of this work is to develop a more reliable technique that can carry out an end-to-end evaluation of adversarial robustness for commercial systems. It requires that this technique can simultaneously deceive black-box recognition models and evade defensive mechanisms. To fulfill this, we design adversarial textured 3D meshes (AT3D) with an elaborate topology on a human face, which can be 3D-printed and pasted on the attacker’s face to evade the defenses. However, the mesh-based optimization regime calculates gradients in high-dimensional mesh space, and can be trapped into local optima with unsatisfactory transferability. To deviate from the mesh-based space, we propose to perturb the low-dimensional coefficient space based on 3D Morphable Model, which significantly improves black-box transferability meanwhile enjoying faster search efficiency and better visual quality. Extensive experiments in digital and physical scenarios show that our method effectively explores the security vulnerabilities of multiple popular commercial services, including three recognition APIs, four anti-spoofing …

Existing face forgery detection models try to discriminate fake images by detecting only spatial artifacts (e.g., generative artifacts, blending) or mainly temporal artifacts (e.g., flickering, discontinuity). They may experience significant performance degradation when facing out-domain artifacts. In this paper, we propose to capture both spatial and temporal artifacts in one model for face forgery detection. A simple idea is to leverage a spatiotemporal model (3D ConvNet). However, we find that it may easily rely on one type of artifact and ignore the other. To address this issue, we present a novel training strategy called AltFreezing for more general face forgery detection. The AltFreezing aims to encourage the model to detect both spatial and temporal artifacts. It divides the weights of a spatiotemporal network into two groups: spatial- and temporal-related. Then the two groups of weights are alternately frozen during the training process so that the model can learn spatial and temporal features to distinguish real or fake videos. Furthermore, we introduce various video-level data augmentation methods to improve the generalization capability of the forgery detection model. Extensive experiments show that our framework outperforms existing methods in terms of generalization to unseen manipulations and datasets.

We introduce an interferometric technique for passive time-of-flight imaging and depth sensing at micrometer axial resolutions. Our technique uses a full-field Michelson interferometer, modified to use sunlight as the only light source. The large spectral bandwidth of sunlight makes it possible to acquire micrometer-resolution time-resolved scene responses, through a simple axial scanning operation. Additionally, the angular bandwidth of sunlight makes it possible to capture time-of-flight measurements insensitive to indirect illumination effects, such as interreflections and subsurface scattering. We build an experimental prototype that we operate outdoors, under direct sunlight, and in adverse environment conditions such as machine vibrations and vehicle traffic. We use this prototype to demonstrate, for the first time, passive imaging capabilities such as micrometer-scale depth sensing robust to indirect illumination, direct-only imaging, and imaging through diffusers.

This paper presents a novel grid-based NeRF called F^2-NeRF (Fast-Free-NeRF) for novel view synthesis, which enables arbitrary input camera trajectories and only costs a few minutes for training. Existing fast grid-based NeRF training frameworks, like Instant-NGP, Plenoxels, DVGO, or TensoRF, are mainly designed for bounded scenes and rely on space warping to handle unbounded scenes. Existing two widely-used space-warping methods are only designed for the forward-facing trajectory or the 360° object-centric trajectory but cannot process arbitrary trajectories. In this paper, we delve deep into the mechanism of space warping to handle unbounded scenes. Based on our analysis, we further propose a novel space-warping method called perspective warping, which allows us to handle arbitrary trajectories in the grid-based NeRF framework. Extensive experiments demonstrate that F^2-NeRF is able to use the same perspective warping to render high-quality images on two standard datasets and a new free trajectory dataset collected by us.

Training a Neural Radiance Field (NeRF) without pre-computed camera poses is challenging. Recent advances in this direction demonstrate the possibility of jointly optimising a NeRF and camera poses in forward-facing scenes. However, these methods still face difficulties during dramatic camera movement. We tackle this challenging problem by incorporating undistorted monocular depth priors. These priors are generated by correcting scale and shift parameters during training, with which we are then able to constrain the relative poses between consecutive frames. This constraint is achieved using our proposed novel loss functions. Experiments on real-world indoor and outdoor scenes show that our method can handle challenging camera trajectories and outperforms existing methods in terms of novel view rendering quality and pose estimation accuracy. Our project page is https://nope-nerf.active.vision.

Neural Radiance Fields (NeRF) have received considerable attention recently, due to its impressive capability in photo-realistic 3D reconstruction and novel view synthesis, given a set of posed camera images. Earlier work usually assumes the input images are of good quality. However, image degradation (e.g. image motion blur in low-light conditions) can easily happen in real-world scenarios, which would further affect the rendering quality of NeRF. In this paper, we present a novel bundle adjusted deblur Neural Radiance Fields (BAD-NeRF), which can be robust to severe motion blurred images and inaccurate camera poses. Our approach models the physical image formation process of a motion blurred image, and jointly learns the parameters of NeRF and recovers the camera motion trajectories during exposure time. In experiments, we show that by directly modeling the real physical image formation process, BAD-NeRF achieves superior performance over prior works on both synthetic and real datasets. Code and data are available at https://github.com/WU-CVGL/BAD-NeRF.

Under good conditions, Neural Radiance Fields (NeRFs) have shown impressive results on novel view synthesis tasks. NeRFs learn a scene’s color and density fields by minimizing the photometric discrepancy between training views and differentiable renderings of the scene. Once trained from a sufficient set of views, NeRFs can generate novel views from arbitrary camera positions. However, the scene geometry and color fields are severely under-constrained, which can lead to artifacts, especially when trained with few input views. To alleviate this problem we learn a prior over scene geometry and color, using a denoising diffusion model (DDM). Our DDM is trained on RGBD patches of the synthetic Hypersim dataset and can be used to predict the gradient of the logarithm of a joint probability distribution of color and depth patches. We show that, these gradients of logarithms of RGBD patch priors serve to regularize geometry and color of a scene. During NeRF training, random RGBD patches are rendered and the estimated gradient of the log-likelihood is backpropagated to the color and density fields. Evaluations on LLFF, the most relevant dataset, show that our learned prior achieves improved quality in the reconstructed geometry and improved generalization to novel views. Evaluations on DTU …

Neural Radiance Field (NeRF) has recently emerged as a powerful representation to synthesize photorealistic novel views. While showing impressive performance, it relies on the availability of dense input views with highly accurate camera poses, thus limiting its application in real-world scenarios. In this work, we introduce Sparse Pose Adjusting Radiance Field (SPARF), to address the challenge of novel-view synthesis given only few wide-baseline input images (as low as 3) with noisy camera poses. Our approach exploits multi-view geometry constraints in order to jointly learn the NeRF and refine the camera poses. By relying on pixel matches extracted between the input views, our multi-view correspondence objective enforces the optimized scene and camera poses to converge to a global and geometrically accurate solution. Our depth consistency loss further encourages the reconstructed scene to be consistent from any viewpoint. Our approach sets a new state of the art in the sparse-view regime on multiple challenging datasets.

Radiance Fields (RF) are popular to represent casually-captured scenes for new view synthesis and several applications beyond it. Mixed reality on personal spaces needs understanding and manipulating scenes represented as RFs, with semantic segmentation of objects as an important step. Prior segmentation efforts show promise but don’t scale to complex objects with diverse appearance. We present the ISRF method to interactively segment objects with fine structure and appearance. Nearest neighbor feature matching using distilled semantic features identifies high-confidence seed regions. Bilateral search in a joint spatio-semantic space grows the region to recover accurate segmentation. We show state-of-the-art results of segmenting objects from RFs and compositing them to another scene, changing appearance, etc., and an interactive segmentation tool that others can use.

Temporal interpolation often plays a crucial role to learn meaningful representations in dynamic scenes. In this paper, we propose a novel method to train spatiotemporal neural radiance fields of dynamic scenes based on temporal interpolation of feature vectors. Two feature interpolation methods are suggested depending on underlying representations, neural networks or grids. In the neural representation, we extract features from space-time inputs via multiple neural network modules and interpolate them based on time frames. The proposed multi-level feature interpolation network effectively captures features of both short-term and long-term time ranges. In the grid representation, space-time features are learned via four-dimensional hash grids, which remarkably reduces training time. The grid representation shows more than 100 times faster training speed than the previous neural-net-based methods while maintaining the rendering quality. Concatenating static and dynamic features and adding a simple smoothness term further improve the performance of our proposed models. Despite the simplicity of the model architectures, our method achieved state-of-the-art performance both in rendering quality for the neural representation and in training speed for the grid representation.

Approximating radiance fields with discretized volumetric grids is one of promising directions for improving NeRFs, represented by methods like DVGO, Plenoxels and TensoRF, which achieve super-fast training convergence and real-time rendering. However, these methods typically require a tremendous storage overhead, costing up to hundreds of megabytes of disk space and runtime memory for a single scene. We address this issue in this paper by introducing a simple yet effective framework, called vector quantized radiance fields (VQRF), for compressing these volume-grid-based radiance fields. We first present a robust and adaptive metric for estimating redundancy in grid models and performing voxel pruning by better exploring intermediate outputs of volumetric rendering. A trainable vector quantization is further proposed to improve the compactness of grid models. In combination with an efficient joint tuning strategy and post-processing, our method can achieve a compression ratio of 100× by reducing the overall model size to 1 MB with negligible loss on visual quality. Extensive experiments demonstrate that the proposed framework is capable of achieving unrivaled performance and well generalization across multiple methods with distinct volumetric structures, facilitating the wide use of volumetric radiance fields methods in real-world applications. Code is available at https://github.com/AlgoHunt/VQRF.

Rendering novel views from captured multi-view images has made considerable progress since the emergence of the neural radiance field. This paper aims to further advance the quality of view rendering by proposing a novel approach dubbed the neural radiance feature field (NRFF). We first propose a multiscale tensor decomposition scheme to organize learnable features so as to represent scenes from coarse to fine scales. We demonstrate many benefits of the proposed multiscale representation, including more accurate scene shape and appearance reconstruction, and faster convergence compared with the single-scale representation. Instead of encoding view directions to model view-dependent effects, we further propose to encode the rendering equation in the feature space by employing the anisotropic spherical Gaussian mixture predicted from the proposed multiscale representation. The proposed NRFF improves state-of-the-art rendering results by over 1 dB in PSNR on both the NeRF and NSVF synthetic datasets. A significant improvement has also been observed on the real-world Tanks & Temples dataset. Code can be found at https://github.com/imkanghan/nrff.

Current 3D scene stylization methods transfer textures and colors as styles using arbitrary style references, lacking meaningful semantic correspondences. We introduce Reference-Based Non-Photorealistic Radiance Fields (Ref-NPR) to address this limitation. This controllable method stylizes a 3D scene using radiance fields with a single stylized 2D view as a reference. We propose a ray registration process based on the stylized reference view to obtain pseudo-ray supervision in novel views. Then we exploit semantic correspondences in content images to fill occluded regions with perceptually similar styles, resulting in non-photorealistic and continuous novel view sequences. Our experimental results demonstrate that Ref-NPR outperforms existing scene and video stylization methods regarding visual quality and semantic correspondence. The code and data are publicly available on the project page at https://ref-npr.github.io.

This paper introduces a novel representation of volumetric videos for real-time view synthesis of dynamic scenes. Recent advances in neural scene representations demonstrate their remarkable capability to model and render complex static scenes, but extending them to represent dynamic scenes is not straightforward due to their slow rendering speed or high storage cost. To solve this problem, our key idea is to represent the radiance field of each frame as a set of shallow MLP networks whose parameters are stored in 2D grids, called MLP maps, and dynamically predicted by a 2D CNN decoder shared by all frames. Representing 3D scenes with shallow MLPs significantly improves the rendering speed, while dynamically predicting MLP parameters with a shared 2D CNN instead of explicitly storing them leads to low storage cost. Experiments show that the proposed approach achieves state-of-the-art rendering quality on the NHR and ZJU-MoCap datasets, while being efficient for real-time rendering with a speed of 41.7 fps for 512 × 512 images on an RTX 3090 GPU. The code is available at https://zju3dv.github.io/mlp_maps/.

Indoor scene reconstruction from monocular images has long been sought after by augmented reality and robotics developers. Recent advances in neural field representations and monocular priors have led to remarkable results in scene-level surface reconstructions. The reliance on Multilayer Perceptrons (MLP), however, significantly limits speed in training and rendering. In this work, we propose to directly use signed distance function (SDF) in sparse voxel block grids for fast and accurate scene reconstruction without MLPs. Our globally sparse and locally dense data structure exploits surfaces’ spatial sparsity, enables cache-friendly queries, and allows direct extensions to multi-modal data such as color and semantic labels. To apply this representation to monocular scene reconstruction, we develop a scale calibration algorithm for fast geometric initialization from monocular depth priors. We apply differentiable volume rendering from this initialization to refine details with fast convergence. We also introduce efficient high-dimensional Continuous Random Fields (CRFs) to further exploit the semantic-geometry consistency between scene objects. Experiments show that our approach is 10× faster in training and 100× faster in rendering while achieving comparable accuracy to state-of-the-art neural implicit methods.

We address the problem of synthesizing novel views from a monocular video depicting a complex dynamic scene. State-of-the-art methods based on temporally varying Neural Radiance Fields (aka dynamic NeRFs) have shown impressive results on this task. However, for long videos with complex object motions and uncontrolled camera trajectories,these methods can produce blurry or inaccurate renderings, hampering their use in real-world applications. Instead of encoding the entire dynamic scene within the weights of MLPs, we present a new approach that addresses these limitations by adopting a volumetric image-based rendering framework that synthesizes new viewpoints by aggregating features from nearby views in a scene motion-aware manner.Our system retains the advantages of prior methods in its ability to model complex scenes and view-dependent effects,but also enables synthesizing photo-realistic novel views from long videos featuring complex scene dynamics with unconstrained camera trajectories. We demonstrate significant improvements over state-of-the-art methods on dynamic scene datasets, and also apply our approach to in-the-wild videos with challenging camera and object motion, where prior methods fail to produce high-quality renderings

Current differentiable renderers provide light transport gradients with respect to arbitrary scene parameters. However, the mere existence of these gradients does not guarantee useful update steps in an optimization. Instead, inverse rendering might not converge due to inherent plateaus, i.e., regions of zero gradient, in the objective function. We propose to alleviate this by convolving the high-dimensional rendering function that maps scene parameters to images with an additional kernel that blurs the parameter space. We describe two Monte Carlo estimators to compute plateau-free gradients efficiently, i.e., with low variance, and show that these translate into net-gains in optimization error and runtime performance. Our approach is a straightforward extension to both black-box and differentiable renderers and enables the successful optimization of problems with intricate light transport, such as caustics or global illumination, that existing differentiable path tracers do not converge on. Our code is at github.com/mfischer-ucl/prdpt.

Inverse rendering methods aim to estimate geometry, materials and illumination from multi-view RGB images. In order to achieve better decomposition, recent approaches attempt to model indirect illuminations reflected from different materials via Spherical Gaussians (SG), which, however, tends to blur the high-frequency reflection details. In this paper, we propose an end-to-end inverse rendering pipeline that decomposes materials and illumination from multi-view images, while considering near-field indirect illumination. In a nutshell, we introduce the Monte Carlo sampling based path tracing and cache the indirect illumination as neural radiance, enabling a physics-faithful and easy-to-optimize inverse rendering method. To enhance efficiency and practicality, we leverage SG to represent the smooth environment illuminations and apply importance sampling techniques. To supervise indirect illuminations from unobserved directions, we develop a novel radiance consistency constraint between implicit neural radiance and path tracing results of unobserved rays along with the joint optimization of materials and illuminations, thus significantly improving the decomposition performance. Extensive experiments demonstrate that our method outperforms the state-of-the-art on multiple synthetic and real datasets, especially in terms of inter-reflection decomposition.

This paper proposes a practical photometric solution for the challenging problem of in-the-wild inverse rendering under unknown ambient lighting. Our system recovers scene geometry and reflectance using only multi-view images captured by a smartphone. The key idea is to exploit smartphone’s built-in flashlight as a minimally controlled light source, and decompose image intensities into two photometric components -- a static appearance corresponds to ambient flux, plus a dynamic reflection induced by the moving flashlight. Our method does not require flash/non-flash images to be captured in pairs. Building on the success of neural light fields, we use an off-the-shelf method to capture the ambient reflections, while the flashlight component enables physically accurate photometric constraints to decouple reflectance and illumination. Compared to existing inverse rendering methods, our setup is applicable to non-darkroom environments yet sidesteps the inherent difficulties of explicit solving ambient reflections. We demonstrate by extensive experiments that our method is easy to implement, casual to set up, and consistently outperforms existing in-the-wild inverse rendering techniques. Finally, our neural reconstruction can be easily exported to PBR textured triangle mesh ready for industrial renderers. Our source code and data are released to https://github.com/za-cheng/WildLight

Human modeling and relighting are two fundamental problems in computer vision and graphics, where high-quality datasets can largely facilitate related research. However, most existing human datasets only provide multi-view human images captured under the same illumination. Although valuable for modeling tasks, they are not readily used in relighting problems. To promote research in both fields, in this paper, we present UltraStage, a new 3D human dataset that contains more than 2,000 high-quality human assets captured under both multi-view and multi-illumination settings. Specifically, for each example, we provide 32 surrounding views illuminated with one white light and two gradient illuminations. In addition to regular multi-view images, gradient illuminations help recover detailed surface normal and spatially-varying material maps, enabling various relighting applications. Inspired by recent advances in neural representation, we further interpret each example into a neural human asset which allows novel view synthesis under arbitrary lighting conditions. We show our neural human assets can achieve extremely high capture performance and are capable of representing fine details such as facial wrinkles and cloth folds. We also validate UltraStage in single image relighting tasks, training neural networks with virtual relighted data from neural assets and demonstrating realistic rendering improvements over prior arts. UltraStage …

We introduce DiffRF, a novel approach for 3D radiance field synthesis based on denoising diffusion probabilistic models. While existing diffusion-based methods operate on images, latent codes, or point cloud data, we are the first to directly generate volumetric radiance fields. To this end, we propose a 3D denoising model which directly operates on an explicit voxel grid representation. However, as radiance fields generated from a set of posed images can be ambiguous and contain artifacts, obtaining ground truth radiance field samples is non-trivial. We address this challenge by pairing the denoising formulation with a rendering loss, enabling our model to learn a deviated prior that favours good image quality instead of trying to replicate fitting errors like floating artifacts. In contrast to 2D-diffusion models, our model learns multi-view consistent priors, enabling free-view synthesis and accurate shape generation. Compared to 3D GANs, our diffusion-based approach naturally enables conditional generation like masked completion or single-view 3D synthesis at inference time.

The subtle vibrations on an object’s surface contain information about the object’s physical properties and its interaction with the environment. Prior works imaged surface vibration to recover the object’s material properties via modal analysis, which discards the transient vibrations propagating immediately after the object is disturbed. Conversely, prior works that captured transient vibrations focused on recovering localized signals (e.g., recording nearby sound sources), neglecting the spatiotemporal relationship between vibrations at different object points. In this paper, we extract information from the transient surface vibrations simultaneously measured at a sparse set of object points using the dual-shutter camera described by Sheinin[31]. We model the geometry of an elastic wave generated shortly after an object’s surface is disturbed (e.g., a knock or a footstep), and use the model to localize the disturbance source for various materials (e.g., wood, plastic, tile). We also show that transient object vibrations contain additional cues about the impact force and the impacting object’s material properties. We demonstrate our approach in applications like localizing the strikes of a ping-pong ball on a table mid-play and recovering the footsteps’ locations by imaging the floor vibrations they create.

We introduce a method that recovers full-surround 3D reconstructions from a single kaleidoscopic image using a neural surface representation. Full-surround 3D reconstruction is critical for many applications, such as augmented and virtual reality. A kaleidoscope, which uses a single camera and multiple mirrors, is a convenient way of achieving full-surround coverage, as it redistributes light directions and thus captures multiple viewpoints in a single image. This enables single-shot and dynamic full-surround 3D reconstruction. However, using a kaleidoscopic image for multi-view stereo is challenging, as we need to decompose the image into multi-view images by identifying which pixel corresponds to which virtual camera, a process we call labeling. To address this challenge, pur approach avoids the need to explicitly estimate labels, but instead “sculpts” a neural surface representation through the careful use of silhouette, background, foreground, and texture information present in the kaleidoscopic image. We demonstrate the advantages of our method in a range of simulated and real experiments, on both static and dynamic scenes.

Learning surface by neural implicit rendering has been a promising way for multi-view reconstruction in recent years. Existing neural surface reconstruction methods, such as NeuS and VolSDF, can produce reliable meshes from multi-view posed images. Although they build a bridge between volume rendering and Signed Distance Function (SDF), the accuracy is still limited. In this paper, we argue that this limited accuracy is due to the bias of their volume rendering strategies, especially when the viewing direction is close to be tangent to the surface. We revise and provide an additional condition for the unbiased volume rendering. Following this analysis, we propose a new rendering method by scaling the SDF field with the angle between the viewing direction and the surface normal vector. Experiments on simulated data indicate that our rendering method reduces the bias of SDF-based volume rendering. Moreover, there still exists non-negligible bias when the learnable standard deviation of SDF is large at early stage, which means that it is hard to supervise the rendered depth with depth priors. Alternatively we supervise zero-level set with surface points obtained from a pre-trained Multi-View Stereo network. We evaluate our method on the DTU dataset and show that it outperforms the …

We present a novel method for reconstructing a 3D implicit surface from a large-scale, sparse, and noisy point cloud. Our approach builds upon the recently introduced Neural Kernel Fields (NKF) representation. It enjoys similar generalization capabilities to NKF, while simultaneously addressing its main limitations: (a) We can scale to large scenes through compactly supported kernel functions, which enable the use of memory-efficient sparse linear solvers. (b) We are robust to noise, through a gradient fitting solve. (c) We minimize training requirements, enabling us to learn from any dataset of dense oriented points, and even mix training data consisting of objects and scenes at different scales. Our method is capable of reconstructing millions of points in a few seconds, and handling very large scenes in an out-of-core fashion. We achieve state-of-the-art results on reconstruction benchmarks consisting of single objects (ShapeNet, ABC), indoor scenes (ScanNet, Matterport3D), and outdoor scenes (CARLA, Waymo).

Masked Modeling (MM) has demonstrated widespread success in various vision challenges, by reconstructing masked visual patches. Yet, applying MM for large-scale 3D scenes remains an open problem due to the data sparsity and scene complexity. The conventional random masking paradigm used in 2D images often causes a high risk of ambiguity when recovering the masked region of 3D scenes. To this end, we propose a novel informative-preserved reconstruction, which explores local statistics to discover and preserve the representative structured points, effectively enhancing the pretext masking task for 3D scene understanding. Integrated with a progressive reconstruction manner, our method can concentrate on modeling regional geometry and enjoy less ambiguity for masked reconstruction. Besides, such scenes with progressive masking ratios can also serve to self-distill their intrinsic spatial consistency, requiring to learn the consistent representations from unmasked areas. By elegantly combining informative-preserved reconstruction on masked areas and consistency self-distillation from unmasked areas, a unified framework called MM-3DScene is yielded. We conduct comprehensive experiments on a host of downstream tasks. The consistent improvement (e.g., +6.1% mAP@0.5 on object detection and +2.2% mIoU on semantic segmentation) demonstrates the superiority of our approach.

Neural Radiance Fields (NeRF) coupled with GANs represent a promising direction in the area of 3D reconstruction from a single view, owing to their ability to efficiently model arbitrary topologies. Recent work in this area, however, has mostly focused on synthetic datasets where exact ground-truth poses are known, and has overlooked pose estimation, which is important for certain downstream applications such as augmented reality (AR) and robotics. We introduce a principled end-to-end reconstruction framework for natural images, where accurate ground-truth poses are not available. Our approach recovers an SDF-parameterized 3D shape, pose, and appearance from a single image of an object, without exploiting multiple views during training. More specifically, we leverage an unconditional 3D-aware generator, to which we apply a hybrid inversion scheme where a model produces a first guess of the solution which is then refined via optimization. Our framework can de-render an image in as few as 10 steps, enabling its use in practical scenarios. We demonstrate state-of-the-art results on a variety of real and synthetic benchmarks.

Existing 3D-aware image synthesis approaches mainly focus on generating a single canonical object and show limited capacity in composing a complex scene containing a variety of objects. This work presents DisCoScene: a 3D-aware generative model for high-quality and controllable scene synthesis. The key ingredient of our method is a very abstract object-level representation (i.e., 3D bounding boxes without semantic annotation) as the scene layout prior, which is simple to obtain, general to describe various scene contents, and yet informative to disentangle objects and background. Moreover, it serves as an intuitive user control for scene editing. Based on such a prior, the proposed model spatially disentangles the whole scene into object-centric generative radiance fields by learning on only 2D images with the global-local discrimination. Our model obtains the generation fidelity and editing flexibility of individual objects while being able to efficiently compose objects and the background into a complete scene. We demonstrate state-of-the-art performance on many scene datasets, including the challenging Waymo outdoor dataset. Our code will be made publicly available.

Triangle mesh segmentation is an important task in 3D shape analysis, especially in applications such as digital humans and AR/VR. Transformer model is inherently permutation-invariant to input, which makes it a suitable candidate model for 3D mesh processing. However, two main challenges involved in adapting Transformer from natural languages to 3D mesh are yet to be solved, such as i) extracting the multi-scale information of mesh data in an adaptive manner; ii) capturing geometric structures of mesh data as the discriminative characteristics of the shape. Current point based Transformer models fail to tackle such challenges and thus provide inferior performance for discretized surface segmentation. In this work, heat diffusion based method is exploited to tackle these problems. A novel Transformer model called MeshFormer is proposed, which i) integrates Heat Diffusion method into Multi-head Self-Attention operation (HDMSA) to adaptively capture the features from local neighborhood to global contexts; ii) applies a novel Heat Kernel Signature based Structure Encoding (HKSSE) to embed the intrinsic geometric structures of mesh instances into Transformer for structure-aware processing. Extensive experiments on triangle mesh segmentation validate the effectiveness of the proposed MeshFormer model and show significant improvements over current state-of-the-art methods.

A key challenge for novel view synthesis of monocular portrait images is 3D consistency under continuous pose variations. Most existing methods rely on 2D generative models which often leads to obvious 3D inconsistency artifacts. We present a 3D-consistent novel view synthesis approach for monocular portrait images based on a recent proposed 3D-aware GAN, namely Generative Radiance Manifolds (GRAM), which has shown strong 3D consistency at multiview image generation of virtual subjects via the radiance manifolds representation. However, simply learning an encoder to map a real image into the latent space of GRAM can only reconstruct coarse radiance manifolds without faithful fine details, while improving the reconstruction fidelity via instance-specific optimization is time-consuming. We introduce a novel detail manifolds reconstructor to learn 3D-consistent fine details on the radiance manifolds from monocular images, and combine them with the coarse radiance manifolds for high-fidelity reconstruction. The 3D priors derived from the coarse radiance manifolds are used to regulate the learned details to ensure reasonable synthesized results at novel views. Trained on in-the-wild 2D images, our method achieves high-fidelity and 3D-consistent portrait synthesis largely outperforming the prior art. Project page: https://yudeng.github.io/GRAMInverter/

We propose pix2pix3D, a 3D-aware conditional generative model for controllable photorealistic image synthesis. Given a 2D label map, such as a segmentation or edge map, our model learns to synthesize a corresponding image from different viewpoints. To enable explicit 3D user control, we extend conditional generative models with neural radiance fields. Given widely-available posed monocular image and label map pairs, our model learns to assign a label to every 3D point in addition to color and density, which enables it to render the image and pixel-aligned label map simultaneously. Finally, we build an interactive system that allows users to edit the label map from different viewpoints and generate outputs accordingly.

We introduce VIVE3D, a novel approach that extends the capabilities of image-based 3D GANs to video editing and is able to represent the input video in an identity-preserving and temporally consistent way. We propose two new building blocks. First, we introduce a novel GAN inversion technique specifically tailored to 3D GANs by jointly embedding multiple frames and optimizing for the camera parameters. Second, besides traditional semantic face edits (e.g. for age and expression), we are the first to demonstrate edits that show novel views of the head enabled by the inherent properties of 3D GANs and our optical flow-guided compositing technique to combine the head with the background video. Our experiments demonstrate that VIVE3D generates high-fidelity face edits at consistent quality from a range of camera viewpoints which are composited with the original video in a temporally and spatially-consistent manner.

In this work, we present a novel framework built to simplify 3D asset generation for amateur users. To enable interactive generation, our method supports a variety of input modalities that can be easily provided by a human, including images, texts, partially observed shapes and combinations of these, further allowing for adjusting the strength of each input. At the core of our approach is an encoder-decoder, compressing 3D shapes into a compact latent representation, upon which a diffusion model is learned. To enable a variety of multi-modal inputs, we employ task-specific encoders with dropout followed by a cross-attention mechanism. Due to its flexibility, our model naturally supports a variety of tasks outperforming prior works on shape completion, image-based 3D reconstruction, and text-to-3D. Most interestingly, our model can combine all these tasks into one swiss-army-knife tool, enabling the user to perform shape generation using incomplete shapes, images, and textual descriptions at the same time, providing the relative weights for each input and facilitating interactivity. Despite our approach being shape-only, we further show an efficient method to texture the generated using large-scale text-to-image models.

Impressive progress in generative models and implicit representations gave rise to methods that can generate 3D shapes of high quality. However, being able to locally control and edit shapes is another essential property that can unlock several content creation applications. Local control can be achieved with part-aware models, but existing methods require 3D supervision and cannot produce textures. In this work, we devise PartNeRF, a novel part-aware generative model for editable 3D shape synthesis that does not require any explicit 3D supervision. Our model generates objects as a set of locally defined NeRFs, augmented with an affine transformation. This enables several editing operations such as applying transformations on parts, mixing parts from different objects etc. To ensure distinct, manipulable parts we enforce a hard assignment of rays to parts that makes sure that the color of each ray is only determined by a single NeRF. As a result, altering one part does not affect the appearance of the others. Evaluations on various ShapeNet categories demonstrate the ability of our model to generate editable 3D objects of improved fidelity, compared to previous part-based generative approaches that require 3D supervision or models relying on NeRFs.

Virtual reality and augmented reality (XR) bring increasing demand for 3D content generation. However, creating high-quality 3D content requires tedious work from a human expert. In this work, we study the challenging task of lifting a single image to a 3D object and, for the first time, demonstrate the ability to generate a plausible 3D object with 360° views that corresponds well with the given reference image. By conditioning on the reference image, our model can fulfill the everlasting curiosity for synthesizing novel views of objects from images. Our technique sheds light on a promising direction of easing the workflows for 3D artists and XR designers. We propose a novel framework, dubbed NeuralLift-360, that utilizes a depth-aware neural radiance representation (NeRF) and learns to craft the scene guided by denoising diffusion models. By introducing a ranking loss, our NeuralLift-360 can be guided with rough depth estimation in the wild. We also adopt a CLIP-guided sampling strategy for the diffusion prior to provide coherent guidance. Extensive experiments demonstrate that our NeuralLift-360 significantly outperforms existing state-of-the-art baselines. Project page: https://vita-group.github.io/NeuralLift-360/

In this paper, we consider the face swapping detection from the perspective of face identity. Face swapping aims to replace the target face with the source face and generate the fake face that the human cannot distinguish between real and fake. We argue that the fake face contains the explicit identity and implicit identity, which respectively corresponds to the identity of the source face and target face during face swapping. Note that the explicit identities of faces can be extracted by regular face recognizers. Particularly, the implicit identity of real face is consistent with the its explicit identity. Thus the difference between explicit and implicit identity of face facilitates face swapping detection. Following this idea, we propose a novel implicit identity driven framework for face swapping detection. Specifically, we design an explicit identity contrast (EIC) loss and an implicit identity exploration (IIE) loss, which supervises a CNN backbone to embed face images into the implicit identity space. Under the guidance of EIC, real samples are pulled closer to their explicit identities, while fake samples are pushed away from their explicit identities. Moreover, IIE is derived from the margin-based classification loss function, which encourages the fake faces with known target identities …

Coordinate-based implicit neural networks, or neural fields, have emerged as useful representations of shape and appearance in 3D computer vision. Despite advances however, it remains challenging to build neural fields for categories of objects without datasets like ShapeNet that provide “canonicalized” object instances that are consistently aligned for their 3D position and orientation (pose). We present Canonical Field Network (CaFi-Net), a self-supervised method to canonicalize the 3D pose of instances from an object category represented as neural fields, specifically neural radiance fields (NeRFs). CaFi-Net directly learns from continuous and noisy radiance fields using a Siamese network architecture that is designed to extract equivariant field features for category-level canonicalization. During inference, our method takes pre-trained neural radiance fields of novel object instances at arbitrary 3D pose, and estimates a canonical field with consistent 3D pose across the entire category. Extensive experiments on a new dataset of 1300 NeRF models across 13 object categories show that our method matches or exceeds the performance of 3D point cloud-based methods.

Recent 3D face reconstruction methods have made significant advances in geometry prediction, yet further cosmetic improvements are limited by lagged albedo because inferring albedo from appearance is an ill-posed problem. Although some existing methods consider prior knowledge from illumination to improve albedo estimation, they still produce a light-skin bias due to racially biased albedo models and limited light constraints. In this paper, we reconsider the relationship between albedo and face attributes and propose an ID2Albedo to directly estimate albedo without constraining illumination. Our key insight is that intrinsic semantic attributes such as race, skin color, and age can constrain the albedo map. We first introduce visual-textual cues and design a semantic loss to supervise facial albedo estimation. Specifically, we pre-define text labels such as race, skin color, age, and wrinkles. Then, we employ the text-image model (CLIP) to compute the similarity between the text and the input image, and assign a pseudo-label to each facial image. We constrain generated albedos in the training phase to have the same attributes as the inputs. In addition, we train a high-quality, unbiased facial albedo generator and utilize the semantic loss to learn the mapping from illumination-robust identity features to the albedo latent codes. …

Synthesizing high-quality 3D face models from natural language descriptions is very valuable for many applications, including avatar creation, virtual reality, and telepresence. However, little research ever tapped into this task. We argue the major obstacle lies in 1) the lack of high-quality 3D face data with descriptive text annotation, and 2) the complex mapping relationship between descriptive language space and shape/appearance space. To solve these problems, we build DESCRIBE3D dataset, the first large-scale dataset with fine-grained text descriptions for text-to-3D face generation task. Then we propose a two-stage framework to first generate a 3D face that matches the concrete descriptions, then optimize the parameters in the 3D shape and texture space with abstract description to refine the 3D face model. Extensive experimental results show that our method can produce a faithful 3D face that conforms to the input descriptions with higher accuracy and quality than previous methods. The code and DESCRIBE3D dataset are released at https://github.com/zhuhao-nju/describe3d.

Sensitivity to severe occlusion and large view angles limits the usage scenarios of the existing monocular 3D dense face alignment methods. The state-of-the-art 3DMM-based method, directly regresses the model’s coefficients, underutilizing the low-level 2D spatial and semantic information, which can actually offer cues for face shape and orientation. In this work, we demonstrate how modeling 3D facial geometry in image and model space jointly can solve the occlusion and view angle problems. Instead of predicting the whole face directly, we regress image space features in the visible facial region by dense prediction first. Subsequently, we predict our model’s coefficients based on the regressed feature of the visible regions, leveraging the prior knowledge of whole face geometry from the morphable models to complete the invisible regions. We further propose a fusion network that combines the advantages of both the image and model space predictions to achieve high robustness and accuracy in unconstrained scenarios. Thanks to the proposed fusion module, our method is robust not only to occlusion and large pitch and roll view angles, which is the benefit of our image space approach, but also to noise and large yaw angles, which is the benefit of our model space method. Comprehensive …

High-fidelity facial avatar reconstruction from a monocular video is a significant research problem in computer graphics and computer vision. Recently, Neural Radiance Field (NeRF) has shown impressive novel view rendering results and has been considered for facial avatar reconstruction. However, the complex facial dynamics and missing 3D information in monocular videos raise significant challenges for faithful facial reconstruction. In this work, we propose a new method for NeRF-based facial avatar reconstruction that utilizes 3D-aware generative prior. Different from existing works that depend on a conditional deformation field for dynamic modeling, we propose to learn a personalized generative prior, which is formulated as a local and low dimensional subspace in the latent space of 3D-GAN. We propose an efficient method to construct the personalized generative prior based on a small set of facial images of a given individual. After learning, it allows for photo-realistic rendering with novel views, and the face reenactment can be realized by performing navigation in the latent space. Our proposed method is applicable for different driven signals, including RGB images, 3DMM coefficients, and audio. Compared with existing works, we obtain superior novel view synthesis results and faithfully face reenactment performance. The code is available here https://github.com/bbaaii/HFA-GP.

Modern 3D-GANs synthesize geometry and texture by training on large-scale datasets with a consistent structure. Training such models on stylized, artistic data, with often unknown, highly variable geometry, and camera information has not yet been shown possible. Can we train a 3D GAN on such artistic data, while maintaining multi-view consistency and texture quality? To this end, we propose an adaptation framework, where the source domain is a pre-trained 3D-GAN, while the target domain is a 2D-GAN trained on artistic datasets. We, then, distill the knowledge from a 2D generator to the source 3D generator. To do that, we first propose an optimization-based method to align the distributions of camera parameters across domains. Second, we propose regularizations necessary to learn high-quality texture, while avoiding degenerate geometric solutions, such as flat shapes. Third, we show a deformation-based technique for modeling exaggerated geometry of artistic domains, enabling---as a byproduct---personalized geometric editing. Finally, we propose a novel inversion method for 3D-GANs linking the latent spaces of the source and the target domains. Our contributions---for the first time---allow for the generation, editing, and animation of personalized artistic 3D avatars on artistic datasets.

This paper presents a 3D diffusion model that automatically generates 3D digital avatars represented as neural radiance fields (NeRFs). A significant challenge for 3D diffusion is that the memory and processing costs are prohibitive for producing high-quality results with rich details. To tackle this problem, we propose the roll-out diffusion network (RODIN), which takes a 3D NeRF model represented as multiple 2D feature maps and rolls out them onto a single 2D feature plane within which we perform 3D-aware diffusion. The RODIN model brings much-needed computational efficiency while preserving the integrity of 3D diffusion by using 3D-aware convolution that attends to projected features in the 2D plane according to their original relationships in 3D. We also use latent conditioning to orchestrate the feature generation with global coherence, leading to high-fidelity avatars and enabling semantic editing based on text prompts. Finally, we use hierarchical synthesis to further enhance details. The 3D avatars generated by our model compare favorably with those produced by existing techniques. We can generate highly detailed avatars with realistic hairstyles and facial hair. We also demonstrate 3D avatar generation from image or text, as well as text-guided editability.

We present Instant Volumetric Head Avatars (INSTA), a novel approach for reconstructing photo-realistic digital avatars instantaneously. INSTA models a dynamic neural radiance field based on neural graphics primitives embedded around a parametric face model. Our pipeline is trained on a single monocular RGB portrait video that observes the subject under different expressions and views. While state-of-the-art methods take up to several days to train an avatar, our method can reconstruct a digital avatar in less than 10 minutes on modern GPU hardware, which is orders of magnitude faster than previous solutions. In addition, it allows for the interactive rendering of novel poses and expressions. By leveraging the geometry prior of the underlying parametric face model, we demonstrate that INSTA extrapolates to unseen poses. In quantitative and qualitative studies on various subjects, INSTA outperforms state-of-the-art methods regarding rendering quality and training time. Project website: https://zielon.github.io/insta/

Over the last few decades, many aspects of human life have been enhanced with virtual domains, from the advent of digital assistants such as Amazon’s Alexa and Apple’s Siri to the latest metaverse efforts of the rebranded Meta. These trends underscore the importance of generating photorealistic visual depictions of humans. This has led to the rapid growth of so-called deepfake and talking-head generation methods in recent years. Despite their impressive results and popularity, they usually lack certain qualitative aspects such as texture quality, lips synchronization, or resolution, and practical aspects such as the ability to run in real-time. To allow for virtual human avatars to be used in practical scenarios, we propose an end-to-end framework for synthesizing high-quality virtual human faces capable of speaking with accurate lip motion with a special emphasis on performance. We introduce a novel network utilizing visemes as an intermediate audio representation and a novel data augmentation strategy employing a hierarchical image synthesis approach that allows disentanglement of the different modalities used to control the global head motion. Our method runs in real-time, and is able to deliver superior results compared to the current state-of-the-art.

We present 3D Cinemagraphy, a new technique that marries 2D image animation with 3D photography. Given a single still image as input, our goal is to generate a video that contains both visual content animation and camera motion. We empirically find that naively combining existing 2D image animation and 3D photography methods leads to obvious artifacts or inconsistent animation. Our key insight is that representing and animating the scene in 3D space offers a natural solution to this task. To this end, we first convert the input image into feature-based layered depth images using predicted depth values, followed by unprojecting them to a feature point cloud. To animate the scene, we perform motion estimation and lift the 2D motion into the 3D scene flow. Finally, to resolve the problem of hole emergence as points move forward, we propose to bidirectionally displace the point cloud as per the scene flow and synthesize novel views by separately projecting them into target image planes and blending the results. Extensive experiments demonstrate the effectiveness of our method. A user study is also conducted to validate the compelling rendering results of our method.

Given two images depicting a person and a garment worn by another person, our goal is to generate a visualization of how the garment might look on the input person. A key challenge is to synthesize a photorealistic detail-preserving visualization of the garment, while warping the garment to accommodate a significant body pose and shape change across the subjects. Previous methods either focus on garment detail preservation without effective pose and shape variation, or allow try-on with the desired shape and pose but lack garment details. In this paper, we propose a diffusion-based architecture that unifies two UNets (referred to as Parallel-UNet), which allows us to preserve garment details and warp the garment for significant pose and body change in a single network. The key ideas behind Parallel-UNet include: 1) garment is warped implicitly via a cross attention mechanism, 2) garment warp and person blend happen as part of a unified process as opposed to a sequence of two separate tasks. Experimental results indicate that TryOnDiffusion achieves state-of-the-art performance both qualitatively and quantitatively.

Predicting natural and diverse 3D hand gestures from the upper body dynamics is a practical yet challenging task in virtual avatar creation. Previous works usually overlook the asymmetric motions between two hands and generate two hands in a holistic manner, leading to unnatural results. In this work, we introduce a novel bilateral hand disentanglement based two-stage 3D hand generation method to achieve natural and diverse 3D hand prediction from body dynamics. In the first stage, we intend to generate natural hand gestures by two hand-disentanglement branches. Considering the asymmetric gestures and motions of two hands, we introduce a Spatial-Residual Memory (SRM) module to model spatial interaction between the body and each hand by residual learning. To enhance the coordination of two hand motions wrt. body dynamics holistically, we then present a Temporal-Motion Memory (TMM) module. TMM can effectively model the temporal association between body dynamics and two hand motions. The second stage is built upon the insight that 3D hand predictions should be non-deterministic given the sequential body postures. Thus, we further diversify our 3D hand predictions based on the initial output from the stage one. Concretely, we propose a Prototypical-Memory Sampling Strategy (PSS) to generate the non-deterministic hand gestures …

Clothes undergo complex geometric deformations, which lead to appearance changes. To edit human videos in a physically plausible way, a texture map must take into account not only the garment transformation induced by the body movements and clothes fitting, but also its 3D fine-grained surface geometry. This poses, however, a new challenge of 3D reconstruction of dynamic clothes from an image or a video. In this paper, we show that it is possible to edit dressed human images and videos without 3D reconstruction. We estimate a geometry aware texture map between the garment region in an image and the texture space, a.k.a, UV map. Our UV map is designed to preserve isometry with respect to the underlying 3D surface by making use of the 3D surface normals predicted from the image. Our approach captures the underlying geometry of the garment in a self-supervised way, requiring no ground truth annotation of UV maps and can be readily extended to predict temporally coherent UV maps. We demonstrate that our method outperforms the state-of-the-art human UV map estimation approaches on both real and synthetic data.

Reconstructing dynamic 3D garment surfaces with open boundaries from monocular videos is an important problem as it provides a practical and low-cost solution for clothes digitization. Recent neural rendering methods achieve high-quality dynamic clothed human reconstruction results from monocular video, but these methods cannot separate the garment surface from the body. Moreover, despite existing garment reconstruction methods based on feature curve representation demonstrating impressive results for garment reconstruction from a single image, they struggle to generate temporally consistent surfaces for the video input. To address the above limitations, in this paper, we formulate this task as an optimization problem of 3D garment feature curves and surface reconstruction from monocular video. We introduce a novel approach, called REC-MV to jointly optimize the explicit feature curves and the implicit signed distance field (SDF) of the garments. Then the open garment meshes can be extracted via garment template registration in the canonical space. Experiments on multiple casually captured datasets show that our approach outperforms existing methods and can produce high-quality dynamic garment surfaces.

We address the problem of clothed human reconstruction from a single image or uncalibrated multi-view images. Existing methods struggle with reconstructing detailed geometry of a clothed human and often require a calibrated setting for multi-view reconstruction. We propose a flexible framework which, by leveraging the parametric SMPL-X model, can take an arbitrary number of input images to reconstruct a clothed human model under an uncalibrated setting. At the core of our framework is our novel self-evolved signed distance field (SeSDF) module which allows the framework to learn to deform the signed distance field (SDF) derived from the fitted SMPL-X model, such that detailed geometry reflecting the actual clothed human can be encoded for better reconstruction. Besides, we propose a simple method for self-calibration of multi-view images via the fitted SMPL-X parameters. This lifts the requirement of tedious manual calibration and largely increases the flexibility of our method. Further, we introduce an effective occlusion-aware feature fusion strategy to account for the most useful features to reconstruct the human model. We thoroughly evaluate our framework on public benchmarks, demonstrating significant superiority over the state-of-the-arts both qualitatively and quantitatively.

We propose a novel approach for unsupervised 3D animation of non-rigid deformable objects. Our method learns the 3D structure and dynamics of objects solely from single-view RGB videos, and can decompose them into semantically meaningful parts that can be tracked and animated. Using a 3D autodecoder framework, paired with a keypoint estimator via a differentiable PnP algorithm, our model learns the underlying object geometry and parts decomposition in an entirely unsupervised manner. This allows it to perform 3D segmentation, 3D keypoint estimation, novel view synthesis, and animation. We primarily evaluate the framework on two video datasets: VoxCeleb 256^2 and TEDXPeople 256^2. In addition, on the Cats 256^2 dataset, we show that it learns compelling 3D geometry even from raw image data. Finally, we show that our model can obtain animatable 3D objects from a singe or a few images.

Over the last few years, with the advent of virtual and augmented reality, an enormous amount of research has been focused on modeling, tracking and reconstructing human hands. Given their power to express human behavior, hands have been a very important, but challenging component of the human body. Currently, most of the state-of-the-art reconstruction and pose estimation methods rely on the low polygon MANO model. Apart from its low polygon count, MANO model was trained with only 31 adult subjects, which not only limits its expressive power but also imposes unnecessary shape reconstruction constraints on pose estimation methods. Moreover, hand appearance remains almost unexplored and neglected from the majority of hand reconstruction methods. In this work, we propose “Handy”, a large-scale model of the human hand, modeling both shape and appearance composed of over 1200 subjects which we make publicly available for the benefit of the research community. In contrast to current models, our proposed hand model was trained on a dataset with large diversity in age, gender, and ethnicity, which tackles the limitations of MANO and accurately reconstructs out-of-distribution samples. In order to create a high quality texture model, we trained a powerful GAN, which preserves high frequency details …

Automated shape repair approaches currently lack access to datasets that describe real-world damaged geometry. We present Fantastic Breaks (and Where to Find Them: https://terascale-all-sensing-research-studio.github.io/FantasticBreaks), a dataset containing scanned, waterproofed, and cleaned 3D meshes for 150 broken objects, paired and geometrically aligned with complete counterparts. Fantastic Breaks contains class and material labels, proxy repair parts that join to broken meshes to generate complete meshes, and manually annotated fracture boundaries. Through a detailed analysis of fracture geometry, we reveal differences between Fantastic Breaks and synthetic fracture datasets generated using geometric and physics-based methods. We show experimental shape repair evaluation with Fantastic Breaks using multiple learning-based approaches pre-trained with synthetic datasets and re-trained with subset of Fantastic Breaks.