Learning Identity-Consistent Feature for Cross-Modality Person Re-Identification via Pixel and Feature Alignment

RGB-IR cross-modality person re-identification (ReID) can be seen as a multicamera retrieval problem that aims to match pedestrian images captured by visible and infrared cameras. Most of the existing methods focus on reducing modality differences through feature representation learning. However, they ignore huge difference in pixel space between two modalities. Unlike these methods, we utilize alignment network (PFANet) reduce modal while aligning features this paper. Our model contains three components, including extractor, generator, joint discriminator. Like previous generator discriminator are used generate high-quality images; however, make substantial improvements extraction module. Firstly, fuse batch normalization global attention (BNG) which pay channel information conducting interaction channels spaces. Secondly, alleviate space, propose mitigation module (MMM). Then, jointly training entire model, our is able not only mitigate intramodality variations but also learn identity-consistent features. Finally, extensive experimental results show outperforms other methods. On SYSU-MM01 dataset, achieves rank-1 accuracy 40.83 % an mAP id="M2"> 39.84 .