MMG-Vid: Maximizing Marginal Gains at Segment-level and Token-level for Efficient Video LLMs
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866909758323163136 |
|---|---|
| author | Ma, Junpeng Zhang, Qizhe Lu, Ming Wang, Zhibin Zhou, Qiang Song, Jun Zhang, Shanghang |
| author_facet | Ma, Junpeng Zhang, Qizhe Lu, Ming Wang, Zhibin Zhou, Qiang Song, Jun Zhang, Shanghang |
| contents | Video Large Language Models (VLLMs) excel in video understanding, but their excessive visual tokens pose a significant computational challenge for real-world applications. Current methods aim to enhance inference efficiency by visual token pruning. However, they do not consider the dynamic characteristics and temporal dependencies of video frames, as they perceive video understanding as a multi-frame task. To address these challenges, we propose MMG-Vid, a novel training-free visual token pruning framework that removes redundancy by Maximizing Marginal Gains at both segment-level and token-level. Specifically, we first divide the video into segments based on frame similarity, and then dynamically allocate the token budget for each segment to maximize the marginal gain of each segment. Subsequently, we propose a temporal-guided DPC algorithm that jointly models inter-frame uniqueness and intra-frame diversity, thereby maximizing the marginal gain of each token. By combining both stages, MMG-Vid can maximize the utilization of the limited token budget, significantly improving efficiency while maintaining strong performance. Extensive experiments demonstrate that MMG-Vid can maintain over 99.5% of the original performance, while effectively reducing 75% visual tokens and accelerating the prefilling stage by 3.9x on LLaVA-OneVision-7B. Code will be released soon. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_21044 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | MMG-Vid: Maximizing Marginal Gains at Segment-level and Token-level for Efficient Video LLMs Ma, Junpeng Zhang, Qizhe Lu, Ming Wang, Zhibin Zhou, Qiang Song, Jun Zhang, Shanghang Computer Vision and Pattern Recognition Video Large Language Models (VLLMs) excel in video understanding, but their excessive visual tokens pose a significant computational challenge for real-world applications. Current methods aim to enhance inference efficiency by visual token pruning. However, they do not consider the dynamic characteristics and temporal dependencies of video frames, as they perceive video understanding as a multi-frame task. To address these challenges, we propose MMG-Vid, a novel training-free visual token pruning framework that removes redundancy by Maximizing Marginal Gains at both segment-level and token-level. Specifically, we first divide the video into segments based on frame similarity, and then dynamically allocate the token budget for each segment to maximize the marginal gain of each segment. Subsequently, we propose a temporal-guided DPC algorithm that jointly models inter-frame uniqueness and intra-frame diversity, thereby maximizing the marginal gain of each token. By combining both stages, MMG-Vid can maximize the utilization of the limited token budget, significantly improving efficiency while maintaining strong performance. Extensive experiments demonstrate that MMG-Vid can maintain over 99.5% of the original performance, while effectively reducing 75% visual tokens and accelerating the prefilling stage by 3.9x on LLaVA-OneVision-7B. Code will be released soon. |
| title | MMG-Vid: Maximizing Marginal Gains at Segment-level and Token-level for Efficient Video LLMs |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2508.21044 |