Unsupervised generalizable multi-source person re-identification: A Domain-specific adaptive framework. (August 2023)

• Record Type: Journal Article
• Authors: Qi, Lei; Liu, Jiaqi; Wang, Lei; Shi, Yinghuan; Geng, Xin
• Is Part Of: Pattern recognition. Volume 140(2023)
• Publication Date: 2023-08
• Subjects: Unsupervised domain generalization person ReID -- Domain-specific adaptive normalization; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2023.109546
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We propose an unsupervised domain generalization person ReID (UDG-ReID) task in the person ReID community. It does not need label information for source domains, and is therefore more challenging but practical than typical DG-ReID task. We develop a simple yet effective domain-specific adaptive framework to reduce the adverse impact of domain gap across source domains during generating pseudo-labels and boost the generalization ability of the model for the unseen target domain. We evaluate our method on multiple benchmark datasets. The results show that it achieves higher accuracy than baselines on all datasets. Besides, our method adapted to supervised DG-ReID task Abstract: Domain generalization (DG) has attracted much attention in person re-identification (ReID) recently. It aims to make a model trained on multiple source domains generalize to an unseen target domain. Although achieving promising progress, existing methods usually need the source domains to be labeled, which could be a significant burden for practical ReID tasks. In this paper, we turn to investigate "unsupervised" domain generalization for ReID, by assuming that no label is available for any source domains. To address this challenging setting, we propose a simple and efficient domain-specific adaptive framework, and realize it with an adaptive normalization module designed upon the batch and instance normalization techniques. In doing so, we successfully yield reliable pseudo-labels to implement training and also enhance the domain generalization capability of the model as required. In addition, we show that our framework can even be applied to improve person ReID under the settings of supervised domain generalization and unsupervised domain adaptation, demonstrating competitive performance with respect to relevant methods. Extensive experimental study on benchmark datasets is conducted to validate the proposed framework. A significance of our work lies in that it shows the potential of unsupervised domain generalization for person ReID and sets a strong baseline for the further research on this topic. The code is available at https://github.com/Qi5Lei/DSAF .