Unsupervised Domain Adaptation for Person Re-identification via Heterogeneous Graph Alignment

Unsupervised person re-identification (re-ID) is becoming increasingly popular due to its power in real-world systems such as public security and intelligent transportation systems. However, the re-ID task challenged by problems of data distribution discrepancy across cameras lack label information. In this paper, we propose a coarse-to-fine heterogeneous graph alignment (HGA) method find cross-camera matches characterizing unlabeled for each camera. coarse-alignment stage, assign projection camera utilize an adversarial learning based align coarse-grained node groups from different into shared space, which consequently alleviates between cameras. fine-alignment exploit potential fine-grained space introduce conservative loss functions constrain aligning process, resulting reliable pseudo labels guidance. The proposed domain adaptation framework not only improves model generalization on target domain, but also facilitates mining integrating discriminative information Extensive experiments benchmark datasets demonstrate that approach outperforms state-of-the-arts.