3DEnhancer: Consistent Multi-View Diffusion for 3D Enhancement

Despite advances in neural rendering, due to the scarcity of high-quality 3D datasets and the inherent limitations of multi-view diffusion models, view synthesis and 3D model generation are restricted to low resolutions with suboptimal multi-view consistency. In this study, we present a novel 3D enhancement pipeline, dubbed 3DEnhancer, which employs a multi-view latent diffusion model to enhance coarse 3D inputs while preserving multi-view consistency. Our method includes a pose-aware encoder and a diffusion-based denoiser to refine low-quality multi-view images, along with data augmentation and multi-view row attention and epipolar aggregation modules to ensure high-quality, consistent 3D outputs across views. Unlike existing video-based approaches, our model supports seamless multi-view enhancement with improved coherence under diverse viewing angles. Extensive evaluations demonstrate that 3DEnhancer significantly outperforms existing methods, improving both multi-view enhancement and per-instance 3D optimization tasks. Code and model will be publicly available.