Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers
Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers
This research introduces a token selection strategy for Visual Geometry Transformers (ViT) to address their quadratic computational cost. The proposed two-stage framework first selects key frames based on diversity and then performs layer-aware sparsification within those frames guided by attention entropy. Experiments show the method accelerates inference by over 85% for scenes with 500 images while maintaining or improving baseline performance, offering a superior speed-accuracy trade-off.
The paper proposes a two-stage token selection framework for Visual Geometry Transformers to reduce quadratic computational costs, achieving over 85% acceleration on scenes with 500 images while maintaining or improving baseline performance.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- code: not_reported
- weights: not_reported
- dataset: not_reported
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
xisting solutions. Extensive experiments show that it accelerates visual geometry transformers by over 85% for scenes with 500 images while maintaining, or even improving, baseline performance, which hints that how our token selection strategy can play a crucial role in future applications of visual geometry transformers. Our project website is available at https://zsh2000.github.io/good-token-hunting.github.io.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org