Improving sparse graph attention for feature matching by informative keypoints exploration

Xingyu Jiang, Shihua Zhang, Xiaoping Zhang, Jiayi Ma

Abstract: Highlights•We propose a novel GNN-based method for fast feature matching.•We introduce a new sparse strategy with emphasized structural cues.•Two designed modules aim to extract structural cues and guide the sparse process.•It performs well in many vision tasks with low computational complexity.