LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding

TL;DR

LongVU introduces a spatiotemporal adaptive compression method that reduces video tokens by leveraging cross-modal queries and inter-frame dependencies, enabling efficient processing of long videos with minimal information loss.

Contribution

The paper presents a novel adaptive compression mechanism for long videos that preserves visual details while significantly reducing token redundancy, improving video understanding performance.

Findings

Outperforms existing methods on various benchmarks.

Effectively processes hour-long videos with minimal information loss.

Scales well with smaller language models.

Abstract

Multimodal Large Language Models (MLLMs) have shown promising progress in understanding and analyzing video content. However, processing long videos remains a significant challenge constrained by LLM's context size. To address this limitation, we propose LongVU, a spatiotemporal adaptive compression mechanism thats reduces the number of video tokens while preserving visual details of long videos. Our idea is based on leveraging cross-modal query and inter-frame dependencies to adaptively reduce temporal and spatial redundancy in videos. Specifically, we leverage DINOv2 features to remove redundant frames that exhibit high similarity. Then we utilize text-guided cross-modal query for selective frame feature reduction. Further, we perform spatial token reduction across frames based on their temporal dependencies. Our adaptive compression strategy effectively processes a large number of frames with little visual information loss within given context length. Our LongVU consistently surpass existing methods across a variety of video understanding benchmarks, especially on hour-long video understanding tasks such as VideoMME and MLVU. Given a light-weight LLM, our LongVU also scales effectively into a smaller size with state-of-the-art video understanding performance.