AutoSSVH: Exploring Automated Frame Sampling for Efficient Self-Supervised Video Hashing

Self-Supervised Video Hashing compresses videos into hash codes for efficient indexing and retrieval by learning meaningful video representations without the need for labeled data. State-of-the-art video hashing methods typically rely on random frame sampling, treating all frames equally. This approach leads to suboptimal hash codes by ignoring frame-specific information density and reconstruction difficulty. To address this limitation, we propose AutoSSVH, a method combining adversarial hard frame mining with hash contrastive learning based on Component Voting. Our adversarial sampling strategy automatically identifies and selects frames with higher reconstruction difficulty, discarding easily reconstructable frames to enhance training rigor and encoding capability. Additionally, we leverage Component Voting in hash contrastive learning, using class-specific anchors and the P2Set paradigm to effectively capture neighborhood information and complex inter-video relationships. Extensive experiments demonstrate that AutoSSVH surpasses existing methods in both accuracy and efficiency. Code and configurations will be released publicly.