Multi-modal person re-identification based on transformer relational regularization

Xiangtian Zheng, Xiaohua Huang, Chen Ji, Xiaolin Yang, Pengcheng Sha, Liang Cheng

Abstract: Highlights•Collaborative module enhances modality, improves stability.•Embedding module enhances features, improves robustness.•Adaptive triplet loss enables adaptive model learning.•TRR achieves optimal VI-ReID performance, validated by studies.