Camera style (camstyle) is a main factor that affects the performance of person re-identification (ReID). In the past years, existing works mainly exploit implicit solutionsfrom the inputs by designing some strong constraints. However, these methods cannotconsistently work as the camstyle still exists in the inputs as well as in the intermediatefeatures. To address this problem, we propose a Camstyle-Identity Disentangling (CID)network for person ReID. More specifically, we disentangle the ID feature and camstylefeature in the latent space. In order to disentangle the features successfully, we presenta Camstyle Shuffling and Retraining (CSR) scheme to generate more ID-preserved andcamstyle variation samples for training. The proposed scheme ensures the success ofdisentangling and is able to eliminate the camstyle features in the backbone during thetraining process. Numerous experimental results on the Market-1501 and DukeMTMC-reID datasets demonstrate that our network can effectively disentangle the features andfacilitate the person ReID networks.
Supplementary material (PDF)