A Cross-Modality Person Re-Identification Method Based on Joint Middle Modality and Representation Learning

Modality differences and intra-class have been hot research problems in the field of cross-modality person re-identification currently. In this paper, we propose a method based on joint middle modality representation learning. To reduce differences, modal generator is used to map different images unified feature space generate images. A two-stream network with parameter sharing extract combined features original image image. addition, multi-granularity pooling strategy combining global local improve learning capability model further differences. optimized by distribution consistency loss, label smoothing cross-entropy hetero-center triplet loss distance accelerate convergence. use publicly available datasets RegDB SYSU-MM01 for validation. The results show that proposed approach paper reaches 68.11% mAP All Search mode dataset 86.54% VtI dataset, performance improvement 3.29% 3.29%, respectively, which demonstrate effectiveness method.