Learning Hybrid Relationships for Person Re-identification

Recently, the relationship among individual pedestrian images and pairwise have become attractive for person re-identification (re-ID) as they effectively improve ability of feature representation. In this paper, we propose a novel method named Hybrid Relationship Network (HRNet) to learn two types relationships in unified framework that makes use their own advantages. Specifically, images, take features nodes construct locally-connected graph, so discriminative nodes. Meanwhile, consistent node constraint inject identity information into graph learning process guide propagate accurately. As treat differences fully-connected estimate robust similarity Furthermore, inter-graph propagation alleviate loss graph. Extensive experiments on Market-1501, DukeMTMCreID, CUHK03 MSMT17 demonstrate proposed HRNet outperforms state-of-the-art methods.