Semi-supervised person re-identification by similarity-embedded cycle GANs
- PDF / 1,697,788 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 90 Downloads / 160 Views
(0123456789().,-volV)(0123456789(). ,- volV)
ORIGINAL ARTICLE
Semi-supervised person re-identification by similarity-embedded cycle GANs Xinyu Zhang1 • Xiao-Yuan Jing1 • Xiaoke Zhu2 • Fei Ma3 Received: 1 October 2019 / Accepted: 21 February 2020 Ó Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Recently, person re-identification (PR-ID) has attracted numerous of research interest because of its broad applications. However, most of the existing PR-ID models always follow the supervised framework, which requires substantial labeled data. In fact, it is often very hard to get enough labeled training samples in many practical application scenarios. To overcome this limitation, some semi-supervised PR-ID methods have been presented more recently. Although some of these semi-supervised models achieve satisfied results, there is still much space to improve. In this paper, we propose a novel semi-supervised PR-ID by similarity-embedded cycle GANs (SECGAN). Our SECGAN model can learn cross-view features with limited labeled data by using cycle GANs. Simultaneously, to further enhance the ability of cycle GANs so that it can extract more discriminative and robust features, similarity metric subnet and specific features extracting subnet are embedded into cycle GANs. Extensive experiments have been conducted on three public PR-ID benchmark datasets, and the experimental results show that our proposed SECGAN approach outperforms several typical supervised methods and the existing state-of-the-art semi-supervised methods including traditional and deep learning semi-supervised methods. Keywords Person re-identification Semi-supervised learning Deep metric learning Similarity embedded Cycle GANs
1 Introduction Nowadays, with the popularization of automated person surveillance and intelligent security, person re-identification (PR-ID) problem has become a hot and important research topic, which attracts many researchers [1–5]. The main task of PR-ID is to identify a designated pedestrian image or picture from a gallery image set obtained from multiple non-overlapping cameras. The largest challenging in the PR-ID task is that there are huge differences among & Xiao-Yuan Jing [email protected] Xinyu Zhang [email protected] 1
School of Computer Science, Wuhan University, Wuhan, China
2
School of Computer and Information Engineering, Henan University, Kaifeng, China
3
School of Computer, Pingdingshan University, Pingdingshan, China
pedestrian images obtained by non-overlapping cameras in appearance, such as body pose, illumination, viewpoint and occlusion. Therefore, PR-ID is a very valuable and challenging task, and it also has high practical value. Based on these reasons, more and more researchers have begun to pay close attention to it. To overcome these difficulties in PR-ID task, many PRID methods have been reported [6–9], and some of them have shown satisfactory results. These PR-Id methods can simply classified into two groups: (i) feature representation-based methods [6, 10]. This kin
Data Loading...
