TKwinFormer: Top k Window Attention in Vision Transformers for Feature Matching

TL;DR

TKwinFormer introduces a novel attention mechanism and multi-stage strategy to enhance local feature matching in vision transformers, effectively integrating global and local information for improved accuracy.

Contribution

The paper proposes Top K Window Attention and a multi-stage matching approach, advancing global-local information integration in vision transformer-based feature matching.

Findings

Outperforms state-of-the-art methods on benchmarks

Improves matching accuracy in low-texture regions

Enhances global information interaction

Abstract

Local feature matching remains a challenging task, primarily due to difficulties in matching sparse keypoints and low-texture regions. The key to solving this problem lies in effectively and accurately integrating global and local information. To achieve this goal, we introduce an innovative local feature matching method called TKwinFormer. Our approach employs a multi-stage matching strategy to optimize the efficiency of information interaction. Furthermore, we propose a novel attention mechanism called Top K Window Attention, which facilitates global information interaction through window tokens prior to patch-level matching, resulting in improved matching accuracy. Additionally, we design an attention block to enhance attention between channels. Experimental results demonstrate that TKwinFormer outperforms state-of-the-art methods on various benchmarks. Code is available at: https://github.com/LiaoYun0x0/TKwinFormer.