technique

visual geometry transformer

techniqueactiveprovisionalvisual-geometry-transformer-939ec986·1 events·first seen 22d ago

Aliases: visual geometry transformer, Visual Geometry Transformers

Co-occurring entities

inter-frame token selection global attention Good Token Hunting multi-view 3D reconstruction intra-frame entropy-guided sparsification intra-frame token sparsification

More like this (12)

ViT (Vision Transformer)Graph Transformer Visual Geometry Foundation Models transformer architecture VisualMem visual language model Transformers transformer-based neural renderer Google Vertex AI Variable-Width Transformers Transformers (library)Geometric Action Model

Recent events (1)

5arXiv · cs.LG·22d ago·source ↗

Good Token Hunting: Token Selection Framework for Visual Geometry Transformers

This paper introduces a two-stage token selection framework to address the quadratic computational scaling of global attention in visual geometry transformers used for multi-view 3D reconstruction. The approach combines diversity-based inter-frame selection (frame-level) with entropy-guided intra-frame sparsification (token-level within frames). Experiments demonstrate over 85% acceleration for 500-image scenes while maintaining or improving baseline reconstruction quality, offering a favorable speed-accuracy trade-off.

Inference Economics Agent and Tool Ecosystem inter-frame token selection visual geometry transformer global attention +5 more