Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers

2026-05-22

AI Summary

This work addresses the scalability limitations of Vision Geometry Transformers, whose computational complexity grows quadratically with sequence length due to global attention. To mitigate this, the authors propose a two-stage token selection strategy: an inter-frame stage that selects keyframes based on scene diversity, and an intra-frame stage that applies layer-aware sparsification through attention entropy to dynamically prune redundant tokens. This approach substantially reduces computational overhead—achieving over 85% acceleration in scenes with 500 images—while maintaining or even slightly improving reconstruction accuracy, thereby offering an excellent trade-off between speed and accuracy.

Abstract

Visual geometry transformers have become powerful architectures for multi-view 3D reconstruction, enabling joint prediction of multiple 3D attributes in a feed-forward manner. However, their computational cost grows quadratically with the input sequence length due to the global attention layers inside these models. This limits both their scalability and efficiency. In this work, we address this challenge with a simple yet general strategy: restricting the number of key/value tokens that each query interacts with during global attention. To achieve effective token selection, we introduce a two-stage framework. First, an inter-frame selection step operates at the frame level to identify frames that should be preserved. Second, an intra-frame selection step further discards more redundant tokens within the selected frames. Our analysis highlights the advantage of a diversity-based strategy for inter-frame selection, which ensures broad coverage of the scene. For intra-frame selection, we show that layer-aware sparsification is necessary, with the selection process guided by the entropy of the global attention pattern. Our approach offers a superior speed-accuracy trade-off compared to existing 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.

Problem

• visual geometry transformers
• computational efficiency
• global attention
• token selection
• 3D reconstruction

Innovation

• token selection
• visual geometry transformers
• sparse attention
• multi-view 3D reconstruction
• efficiency-accuracy trade-off