Visible-Infrared Pedestrian Re-Identification via Heterogeneity Augmented Cross-Modality Learning

Visible-Infrared Cross Modality pedestrian Re-identification (CmRe-id) focuses on mapping pedestrian photographs from different cameras of the same identity. Despite significant achievements in previous works, it remains restriction: cross-modality feature mismatch caused by data heterogeneity. To overcome this restriction, in this paper, a heterogeneity augmented cross-modality learning is proposed to handle feature representations among modalities. Specifically, to cope with cross-modality feature mismatch, we design a Heterogeneity Augmented Module. It utilizes channel attention mechanism, extract heterogenous features of different modalities from feature maps. With channel attention mechanism, the HAM can better perceive identity information on both modalities. Comprehensive experiments on the public SYSU-MM01 and RegDB datasets reveal the proposed method’s superiority over current practices.