Exploiting Sample Uncertainty for Domain Adaptive Person Re-Identification

• DOI: 10.1609/aaai.v35i4.16468
• Publication Date: 2022-09-08T18:16:49Z
• Authors: Kecheng Zheng; Cuiling Lan; Wenjun Zeng; Zhizheng Zhang; Zheng-Jun Zha
• Keywords: Benchmark (surveying); Sample (material); Feature (linguistics); Representation

Many unsupervised domain adaptive (UDA) person ReID approaches combine clustering-based pseudo-label prediction with feature fine-tuning. However, because of gap, the pseudo-labels are not always reliable and there noisy/incorrect labels. This would mislead representation learning deteriorate performance. In this paper, we propose to estimate exploit credibility assigned each sample alleviate influence noisy labels, by suppressing contribution samples. We build our baseline framework using mean teacher method together an additional contrastive loss. have observed that a wrong through clustering in general has weaker consistency between output model student model. Based on finding, uncertainty (measured levels) evaluate reliability incorporate re-weight its within various losses, including ID classification loss per sample, triplet loss, Our uncertainty-guided optimization brings significant improvement achieves state-of-the-art performance benchmark datasets.