Dual Distribution Alignment Network for Generalizable Person Re-Identification

Domain generalization (DG) is promising to handle person Re-Identification (Re-ID), which trains the model using labels from the source domain alone, and then directly adopts the trained model to the target domain without model updating. However, existing DG approaches are still defected when facing serious domain variations. Therefore, DG highly relies on designing domain-invariant features, which is still an open problem, since most existing approaches directly mix multiple datasets to train DG models without considering the inter-domain similarities, i.e., examples that are very similar but from different domains. In this paper, we present a Dual Distribution Alignment Network (DDAN), which maps images into a domain-invariant feature space by selectively aligning the distributions of multiple source domains. To this end, an alignment network is designed with dual-level constraints, i.e., a novel domainwise adversarial feature learning and an identity-wise similarity enhancement. We evaluate our DDAN on a large-scale Domain Generalization Re-ID (DG Re-ID) benchmark. Quantitative results demonstrate that the proposed DDAN can well align the distributions of multiple domains with serious variations, and significantly outperform all existing domain generalization approaches. CCS CONCEPTS • Information systems → Information retrieval.