Progressive Unsupervised Person Re-Identification by Tracklet Association With Spatio-Temporal Regularization

Existing methods for person re-identification (Re-ID) are mostly based on supervised learning which requires numerous manually labeled samples across all camera views training. Such a paradigm suffers the scalability issue since in real-world Re-ID application, it is difficult to exhaustively label abundant identities over multiple disjoint views. To this end, we propose progressive deep method unsupervised wild by Tracklet Association with Spatio-Temporal Regularization (TASTR). In our approach, first collect tracklet data within each automatic detection and tracking. Then, an initial model trained within-camera triplet construction representation learning. After that, visual feature spatio-temporal constraint, associate cross-camera tracklets generate triplets update model. Lastly, refined model, better of can be extracted, further promote association tracklets. The last two steps iterated times progressively upgrade facilitate study, have collected new 4K UHD video dataset named Campus4K full frames information. Experimental results show that constraint training phase, proposed approach outperforms state-of-the-art notable margins DukeMTMC-reID, achieves competitive performance fully both DukeMTMC-reID datasets.