作者：Shen Zehong, Bao Hujun, Zhou Xiaowei；等
来源：IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
卷:45 期:6 页:7726-7738 出版时间:JUN 2023
We propose a novel framework named LoFTR for local image feature matching, which is a fundamental problem in computer vision. Unlike traditional methods that perform feature detection, description, and matching sequentially, LoFTR proposes to establish pixel-wise dense matches in a coarse-to-fine manner using Transformers, getting rid of the need of feature detection and largely improving matching robustness for challenging scenarios such as low-texture regions and large illumination/viewpoint changes.
