DynTok: Dynamic Compression of Visual Tokens for Efficient and Effective Video Understanding
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913875105939456 |
|---|---|
| author | Zhang, Hongzhi Zhang, Jingyuan Ji, Xingguang Wang, Qi Zhang, Fuzheng |
| author_facet | Zhang, Hongzhi Zhang, Jingyuan Ji, Xingguang Wang, Qi Zhang, Fuzheng |
| contents | Typical video modeling methods, such as LLava, represent videos as sequences of visual tokens, which are then processed by the LLM backbone for effective video understanding. However, this approach leads to a massive number of visual tokens, especially for long videos. A practical solution is to first extract relevant visual information from the large visual context before feeding it into the LLM backbone, thereby reducing computational overhead. In this work, we introduce DynTok, a novel \textbf{Dyn}amic video \textbf{Tok}en compression strategy. DynTok adaptively splits visual tokens into groups and merges them within each group, achieving high compression in regions with low information density while preserving essential content. Our method reduces the number of tokens to 44.4% of the original size while maintaining comparable performance. It further benefits from increasing the number of video frames and achieves 65.3% on Video-MME and 72.5% on MLVU. By applying this simple yet effective compression method, we expose the redundancy in video token representations and offer insights for designing more efficient video modeling techniques. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_03990 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | DynTok: Dynamic Compression of Visual Tokens for Efficient and Effective Video Understanding Zhang, Hongzhi Zhang, Jingyuan Ji, Xingguang Wang, Qi Zhang, Fuzheng Computation and Language Computer Vision and Pattern Recognition Typical video modeling methods, such as LLava, represent videos as sequences of visual tokens, which are then processed by the LLM backbone for effective video understanding. However, this approach leads to a massive number of visual tokens, especially for long videos. A practical solution is to first extract relevant visual information from the large visual context before feeding it into the LLM backbone, thereby reducing computational overhead. In this work, we introduce DynTok, a novel \textbf{Dyn}amic video \textbf{Tok}en compression strategy. DynTok adaptively splits visual tokens into groups and merges them within each group, achieving high compression in regions with low information density while preserving essential content. Our method reduces the number of tokens to 44.4% of the original size while maintaining comparable performance. It further benefits from increasing the number of video frames and achieves 65.3% on Video-MME and 72.5% on MLVU. By applying this simple yet effective compression method, we expose the redundancy in video token representations and offer insights for designing more efficient video modeling techniques. |
| title | DynTok: Dynamic Compression of Visual Tokens for Efficient and Effective Video Understanding |
| topic | Computation and Language Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2506.03990 |