Disentangling Reconstruction Network for Unsupervised Cross-Domain Person Re-Identification

Abstract

Unsupervised cross-domain Person Re-Identification (Re-ID) suffers from severe domain gap issue. While different works address this issue, bridging domain gap with high-level representation is hard as it comprises of entangled information including identity, background, occlusion, and other domain-specific variations. In this paper, we propose a disentangled reconstruction method to address the domain-shift problem for Re-ID in an unsupervised manner. To this end, we have two major contributions. First, we propose to disentangle identity-relevant and identity-irrelevant features from person images. Second, in the target domain, we explicitly consider the camera style transfer images as a data augmentation to address intra-domain discrepancy and to learn the camera invariant features. Experimental results on the challenging benchmarks of Market-1501 and DukeMTMC-reID demonstrate that our proposed method achieves competitive performance.