The devil in the tail: Cluster consolidation plus cluster adaptive balancing loss for unsupervised person re-identification. (September 2022)

• Record Type: Journal Article
• Authors: Li, Mingkun; Sun, He; Lin, Chaoqun; Li, Chun-Guang; Guo, Jun
• Is Part Of: Pattern recognition. Volume 129(2022)
• Publication Date: 2022-09
• Subjects: Unsupervised person re-identification -- Cluster consolidation -- Cluster adaptive balancing loss -- Long-tail problem; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108763
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We propose a simple yet effective approach, called cluster consolidation (CC), to reorganize the clustering result. The reorganization step can improve the compactness of larger clusters by pruning a proportion of unreliable samples into tiny clusters or singletons. We propose a cluster adaptive balancing (CAB) loss to effectively train the network by automatically assigning proper weights to the imbalanced and noisy pseudo labels. In this way, the unsupervised person Re-ID task is formulated as a cluster adaptive long-tail learning problem. Extensive experiments on widely used benchmark datasets are conducted and demonstrate state-of-the-art performance. A set of ablation studies are also provided. Abstract: Unsupervised person re-identification (Re-ID) is to retrieve pedestrians from different camera views without supervision information. State-of-the-art methods are usually built upon training a convolution neural network with pseudo labels generated by clustering. Unfortunately, the pseudo labels are highly unbalanced and heavily noisy, carrying ineffective or even erroneous supervision information. To address these deficiencies, we present an effective clustering and reorganization approach, called Cluster Consolidation, which aims to separate a small proportion of unreliable data points from each cluster. This approach benefits to improve the quality of the pseudo labels, but also yields more tiny clusters. Thus, we further propose a Cluster Adaptive Balancing (CAB) loss to effectively train the network with the imbalance pseudo labels, where our CAB loss is able to automatically balance the importance of each cluster. We conduct extensive experiments on widely used person Re-ID benchmark datasets and demonstrate the effectiveness of our proposals.