Attention Network Analysis for Image Feature Matching

ChoHee Kim, Timothy Woinoski, Ivan V. Bajic, Seong-Won Lee

Abstract: Feature detection and feature matching have long been central tasks in the field of computer vision, with extensive research being conducted over the years. Recently, various models utilizing attention mechanisms have demonstrated outstanding performance, leading to the proposal of numerous methods that use attention to interpret images in vision tasks. In this study, we aim to propose an attention structure that effectively extracts features from image patterns, making it suitable for various vision applications. To achieve this, we conduct both quantitative and qualitative evaluations of the performance of different attention structures, providing a comprehensive analysis of how well these mechanisms are suited to specific tasks. The analysis will serve as a crucial guideline for achieving optimal performance in various computer vision applications.