MSJoE: Jointly Evolving MLLM and Sampler for Efficient Long-Form Video Understanding

TL;DR

MSJoE introduces a joint framework that combines evolving large language models with a lightweight key-frame sampler, significantly improving long-form video understanding efficiency and accuracy.

Contribution

The paper proposes a novel joint evolution framework for MLLMs and key-frame samplers, optimizing both through reinforcement learning for better long-video comprehension.

Findings

Achieves 8.0% accuracy improvement over base MLLM

Outperforms strongest baseline by 1.1% accuracy

Demonstrates effectiveness on multiple long-video QA datasets

Abstract

Efficiently understanding long-form videos remains a fundamental challenge for multimodal large language models (MLLMs). In this paper, we present MLLM-Sampler Joint Evolution (MSJoE), a novel framework that jointly evolves the MLLM and a lightweight key-frame sampler for efficient long-form video understanding. MSJoE builds upon a key assumption that only a small subset of key-frames is truly informative for answering each question to a video. Specifically, MSJoE first reasons out several queries, which describe diverse visual perspectives relevant to the question. Then, these queries interact with a frozen CLIP model to produce a query-frame similarity matrix. Finally, a lightweight sampler predicts key-frame sampling weights from this matrix, selecting a compact set of informative frames, which are then fed into the MLLM for answer generation. Both the MLLM and sampler are jointly optimized through reinforcement learning, enabling co-adaptation of query-reasoning, frame-sampling, and key-frame understanding. A new long-video QA dataset containing 2.8K videos with 7K question-answer pairs is collected to support the training process. Extensive experiments on VideoMME, LongVideoBench, LVBench, and MLVU show that MSJoE achieves 8.0\% accuracy gain upon the base MLLM, and 1.1\% higher accuracy than strongest baseline method.