DyMU: Dynamic Merging and Virtual Unmerging for Efficient VLMs
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_ | 1866909606960168960 |
|---|---|
| author | Wang, Zhenhailong Purushwalkam, Senthil Xiong, Caiming Savarese, Silvio Ji, Heng Xu, Ran |
| author_facet | Wang, Zhenhailong Purushwalkam, Senthil Xiong, Caiming Savarese, Silvio Ji, Heng Xu, Ran |
| contents | We present DyMU, an efficient, training-free framework that dynamically reduces the computational burden of vision-language models (VLMs) while maintaining high task performance. Our approach comprises two key components. First, Dynamic Token Merging (DToMe) reduces the number of visual token embeddings by merging similar tokens based on image complexity, addressing the inherent inefficiency of fixed-length outputs in vision transformers. Second, Virtual Token Unmerging (VTU) simulates the expected token sequence for large language models (LLMs) by efficiently reconstructing the attention dynamics of a full sequence, thus preserving the downstream performance without additional fine-tuning. Unlike previous approaches, our method dynamically adapts token compression to the content of the image and operates completely training-free, making it readily applicable to most state-of-the-art VLM architectures. Extensive experiments on image and video understanding tasks demonstrate that DyMU can reduce the average visual token count by 32%-85% while achieving comparable performance to full-length models across diverse VLM architectures, including the recently popularized AnyRes-based visual encoders. Furthermore, through qualitative analyses, we demonstrate that DToMe effectively adapts token reduction based on image complexity and, unlike existing systems, provides users more control over computational costs. Project page: https://mikewangwzhl.github.io/dymu/. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_17040 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | DyMU: Dynamic Merging and Virtual Unmerging for Efficient VLMs Wang, Zhenhailong Purushwalkam, Senthil Xiong, Caiming Savarese, Silvio Ji, Heng Xu, Ran Computer Vision and Pattern Recognition Artificial Intelligence We present DyMU, an efficient, training-free framework that dynamically reduces the computational burden of vision-language models (VLMs) while maintaining high task performance. Our approach comprises two key components. First, Dynamic Token Merging (DToMe) reduces the number of visual token embeddings by merging similar tokens based on image complexity, addressing the inherent inefficiency of fixed-length outputs in vision transformers. Second, Virtual Token Unmerging (VTU) simulates the expected token sequence for large language models (LLMs) by efficiently reconstructing the attention dynamics of a full sequence, thus preserving the downstream performance without additional fine-tuning. Unlike previous approaches, our method dynamically adapts token compression to the content of the image and operates completely training-free, making it readily applicable to most state-of-the-art VLM architectures. Extensive experiments on image and video understanding tasks demonstrate that DyMU can reduce the average visual token count by 32%-85% while achieving comparable performance to full-length models across diverse VLM architectures, including the recently popularized AnyRes-based visual encoders. Furthermore, through qualitative analyses, we demonstrate that DToMe effectively adapts token reduction based on image complexity and, unlike existing systems, provides users more control over computational costs. Project page: https://mikewangwzhl.github.io/dymu/. |
| title | DyMU: Dynamic Merging and Virtual Unmerging for Efficient VLMs |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2504.17040 |