Structured Domain Adaptation With Online Relation Regularization for Unsupervised Person Re-ID

Ge, Yixiao; Zhu, Feng; Chen, Dapeng; Zhao, Rui; Wang, Xiaogang; Li, Hongsheng

doi:10.1109/tnnls.2022.3173489

Cited by 53 publications

(43 citation statements)

References 75 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Unsupervised domain adaptation methods utilize transfer learning to improve the person re-ID performance on the target domain. Existing unsupervised domain adaptation methods can further be divided into two main categories: pseudo label-based methods [ 4 , 12 , 13 , 14 , 15 , 16 , 17 , 18 ] and domain translation-based methods [ 19 , 20 , 21 , 22 , 23 ]. Pseudo label-based methods first pre-train the model on the source domain and extract the features of the instances in the target domain based on the pre-trained model.…”

Section: Related Workmentioning

confidence: 99%

Discriminatively Unsupervised Learning Person Re-Identification via Considering Complicated Images

Liang

2023

Sensors

View full text Add to dashboard Cite

State-of-the-art purely unsupervised learning person re-ID methods first cluster all the images into multiple clusters and assign each clustered image a pseudo label based on the cluster result. Then, they construct a memory dictionary that stores all the clustered images, and subsequently train the feature extraction network based on this dictionary. All these methods directly discard the unclustered outliers in the clustering process and train the network only based on the clustered images. The unclustered outliers are complicated images containing different clothes and poses, with low resolution, severe occlusion, and so on, which are common in real-world applications. Therefore, models trained only on clustered images will be less robust and unable to handle complicated images. We construct a memory dictionary that considers complicated images consisting of both clustered and unclustered images, and design a corresponding contrastive loss by considering both kinds of images. The experimental results show that our memory dictionary that considers complicated images and contrastive loss can improve the person re-ID performance, which demonstrates the effectiveness of considering unclustered complicated images in unsupervised person re-ID.

show abstract

Section: Related Workmentioning

confidence: 99%

Discriminatively Unsupervised Learning Person Re-Identification via Considering Complicated Images

Liang

2023

Sensors

View full text Add to dashboard Cite

show abstract

“…For example, BUC [34] treats each individual image as an identity and applies a bottom-up clustering to the feature embedding reducing the number of classes. SPCL [35] proposes a self-paced contrastive learning strategy with a novel clustering reliability criterion to generate more reliable clusters. Some methods utilize cross-domain transfer learning [36], [37].…”

Section: B Unsupervised Person Re-idmentioning

confidence: 99%

“…For intra-image contrastive learning, we present a novel deep intraimage contrastive learning strategy that includes a spatialinvariant contrast module and an occlusion-invariant contrast module, described below. For memory-based contrastive learning, we build a cluster-level re-id memory bank using SPCL [35], and supervise the re-id feature learning like [11]. The memory bank is initialized with the averaged re-id features of two branches.…”

Section: B Overall Architecturementioning

confidence: 99%

See 1 more Smart Citation

Deep Intra-Image Contrastive Learning for Weakly Supervised One-Step Person Search

Wang¹,

Pang²,

Cao³

et al. 2023

Preprint

View full text Add to dashboard Cite

Weakly supervised person search aims to perform joint pedestrian detection and re-identification (re-id) with only person bounding-box annotations. Recently, the idea of contrastive learning is initially applied to weakly supervised person search, where two common contrast strategies are memorybased contrast and intra-image contrast. We argue that current intra-image contrast is shallow, which suffers from spatiallevel and occlusion-level variance. In this paper, we present a novel deep intra-image contrastive learning using a Siamese network. Two key modules are spatial-invariant contrast (SIC) and occlusion-invariant contrast (OIC). SIC performs many-toone contrasts between two branches of Siamese network and dense prediction contrasts in one branch of Siamese network. With these many-to-one and dense contrasts, SIC tends to learn discriminative scale-invariant and location-invariant features to solve spatial-level variance. OIC enhances feature consistency with the masking strategy to learn occlusion-invariant features. Extensive experiments are performed on two person search datasets CUHK-SYSU and PRW, respectively. Our method achieves a state-of-the-art performance among weakly supervised one-step person search approaches. We hope that our simple intra-image contrastive learning can provide more paradigms on weakly supervised person search. The source code is available at https://github.com/jiabeiwangTJU/DICL.

show abstract

“…There are two main categories of the existing UDA methods for target recognition. One is the methods including pseudo label [28–30], and the other is domain transformation methods [2, 31, 32]. Deng et al.…”

Section: Related Workmentioning

confidence: 99%

An unsupervised person re‐identification approach based on cross‐view distribution alignment

Jia

Wang

2021

IET Image Processing

View full text Add to dashboard Cite

Unsupervised clustering is a kind of popular solution for unsupervised person reidentification (re-ID). However, due to the influence of cross-view differences, the results of clustering labels are not accurate. To solve this problem, an unsupervised re ID method based on cross-view distributed alignment (CV-DA) to reduce the influence of unsupervised cross-view is proposed. Specifically, based on a popular unsupervised clustering method, density clustering DBSCAN is used to obtain pseudo labels. By calculating the similarity scores of images in the target domain and the source domain, the similarity distribution of different camera views is obtained and is aligned with the distribution with the consistency constraint of pseudo labels. The cross-view distribution alignment constraint is used to guide the clustering process to obtain a more reliable pseudo label. The comprehensive comparative experiments are done in two public datasets, i.e. Market-1501 and DukeMTMC-reID. The comparative results show that the proposed method outperforms several state-of-the-art approaches with mAP reaching 52.6% and rank1 71.1%. In order to prove the effectiveness of the proposed CV-DA, the proposed constraint is added into two advanced re-ID methods. The experimental results demonstrate that the mAP and rank increase by ∽0.5-2% after using the cross-view distribution alignment constraint comparing with that of the associated original methods without using CV-DA. INTRODUCTIONPerson re-identification (re-ID) [1] aims to locate the target person in surveillance videos with a given probe image. With the rapid development of deep learning to techniques, most person re-ID methods focus on supervised algorithm and have achieved high accuracy on public datasets. However, they need to have many pairs of label data between each pair of camera views, which limits the scalability of large-scale applications. In large-scale programs, manually marking pairs of re-ID data is time-consuming and laborious, only unlabelled data is available. In order to solve the scalability problem of re-ID large-scale program, unsupervised re-ID has been widely concerned. Unsupervised learning is achieved by migrating the source domain to the target domain. However, due to the data deviation between the source dataset and the target dataset, the target domain's performance may decrease significantly [2,3].This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

show abstract

Structured Domain Adaptation With Online Relation Regularization for Unsupervised Person Re-ID

Cited by 53 publications

References 75 publications

Discriminatively Unsupervised Learning Person Re-Identification via Considering Complicated Images

Discriminatively Unsupervised Learning Person Re-Identification via Considering Complicated Images

Deep Intra-Image Contrastive Learning for Weakly Supervised One-Step Person Search

An unsupervised person re‐identification approach based on cross‐view distribution alignment

Contact Info

Product

Resources

About