$\infty$-Video: A Training-Free Approach to Long Video Understanding via Continuous-Time Memory Consolidation
This work tackles the problem of efficient and scalable long-video understanding for users of video-language models, offering an incremental improvement by extending existing architectures without retraining.
This paper addresses the challenge of long-video understanding in video-language models, which are limited by context lengths and sparse frame subsampling. The proposed $\infty$-Video framework enables processing of arbitrarily long videos by integrating a continuous-time long-term memory consolidation mechanism into existing video Q-formers, improving performance in video question-answering tasks with Video-LLaMA and VideoChat2.
Current video-language models struggle with long-video understanding due to limited context lengths and reliance on sparse frame subsampling, often leading to information loss. This paper introduces $\infty$-Video, which can process arbitrarily long videos through a continuous-time long-term memory (LTM) consolidation mechanism. Our framework augments video Q-formers by allowing them to process unbounded video contexts efficiently and without requiring additional training. Through continuous attention, our approach dynamically allocates higher granularity to the most relevant video segments, forming "sticky" memories that evolve over time. Experiments with Video-LLaMA and VideoChat2 demonstrate improved performance in video question-answering tasks, showcasing the potential of continuous-time LTM mechanisms to enable scalable and training-free comprehension of long videos.