Abstract:Compared to traditional supervised and and unsupervised learning methods，one-shot person re-identification has emerged as a research focus due to its practical applicability. Despite significant advancements in this field in recent years，the performance of existing methods is still limited by several challenges such as label noise interference and restricted pseudo-label confidence. This paper proposes a label confidence ranking guided multilateral learning algorithm to overcome them. Firstly，a label confidence ranking strategy based on the hierarchical clustering module is proposed to classify unlabeled image instances and exclude noises. Secondly，a multilateral learning framework is introduced by leveraging the label confidence ranking to conduct training in multiple network branches. Finally，through extensive experiments on three public benchmark datasets，Market-1501，DukeMTMC-reID and CUHK03，it is verified that the proposed method can outperform most present algorithms with good performance on both mAP and Rank-k with robustness.