Recurrent Attention-based Token Selection for Efficient Streaming Video-LLMs
This addresses efficiency issues for real-time video understanding in applications like surveillance or live streaming, though it is incremental as it builds on existing Video-LLMs.
The paper tackles the challenge of processing hour-long videos in streaming scenarios for Video-LLMs by proposing a training-free method that selects important visual tokens based on LLM attention, discarding up to ~95% of tokens with minimal performance loss, and achieves state-of-the-art performance on streaming video benchmarks.
Video Large Language Models (Video-LLMs) excel at understanding videos in-context, provided they have full access to the video when answering queries. However, these models face challenges in streaming scenarios where hour-long videos must be processed online, and questions need timely responses. In this work, we propose a training-free approach compatible with standard Video-LLMs, leveraging three key concepts: 1) LLM-informed selection of visual tokens to identify those that the LLM has attended to and contributed to its understanding of each short clip. Our attention-based selection allows us to discard up to ~95% of unimportant visual tokens with minimal performance loss; 2) Recurrent processing of past selected tokens to generate temporally coherent understanding of each processed clip; 3) Caption-based question answering for lightweight and accurate responses. Our method achieves state-of-the-art performance on streaming video benchmarks, striking a balance between efficiency and effectiveness.