This work considers the problem of unsupervised domain adaptation in person
re-identification (re-ID), which aims to transfer knowledge from the source
domain to the target domain. Existing methods are primary to reduce the
inter-domain shift between the domains, which however usually