Unsupervised Person Re-Identification via Multi-Label Classification

The challenge of unsupervised person re-identification (ReID) lies in learning discriminative features without true labels. Most previous works predict single-class pseudo labels through clustering. To improve the quality generated labels, this paper formulates ReID as a multi-label classification task to progressively seek Our method starts by assigning each image with label, then evolves leveraging updated model for label prediction. We first investigate effect precision and recall rates accuracy. This study motivates Clustering-guided Multi-class Label Prediction (CMLP), which adopts clustering cycle consistency ensure high rate reasonably good boost efficiency, we further propose Memory-based Multi-label Classification Loss (MMCL). MMCL memory-based non-parametric classifier integrates local loss global optimization efficiency. CMLP work iteratively substantially performance. Experiments on several large-scale datasets demonstrate superiority our ReID. For instance, fully setting achieve rank-1 accuracy 90.1% Market-1501, already outperforming many transfer supervised methods.