Deep Transfer Learning for Person Re-identification

Geng, Mengyue; Wang, Yaowei; Xiang, Tao; Tian, Yonghong

doi:10.48550/arxiv.1611.05244

Cited by 99 publications

(100 citation statements)

References 54 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In order to address these issues, most existing person Re-ID methods are designed on supervised learning [45,46,28,14,47,43,44,19] which aims to learn a dis-criminative representation from labeled data. Recently, benefited from the success of deep learning [20,6,18], the supervised learning based methods have obtained significant performance improvement [35,7].…”

Section: Market-1501mentioning

confidence: 99%

Domain Adaptive Attention Model for Unsupervised Cross-Domain Person Re-Identification

Huang,

Peng,

Jin

et al. 2019

Preprint

View full text Add to dashboard Cite

Person re-identification (Re-ID) across multiple datasets is a challenging yet important task due to the possibly large distinctions between different datasets and the lack of labeled training samples in practical applications. This work proposes a novel unsupervised domain adaption framework which transfers the discriminative representations from the labeled source domain (dataset) to the unlabeled target domain (dataset). Given the feature map of any image from a backbone network, a novel domain adaptive attention model (DAAM) is proposed to separate the feature map into the domain-shared feature (DSH) map and the domainspecific (DSP) feature map simultaneously. The former is transferable and used to help the Re-ID task in the target domain, while the later is modeled to alleviate negative transfer caused by the domain divergence. Then, a DSH branch and a DSP branch are introduced to the network to learn these two feature maps respectively. In addition, we propose to formulate the domain adaption task as an one-class classification problem with a novel domain similarity loss, and a novel person Re-ID loss is proposed to take full use of unlabeled target data by estimating weak labels and weighted cross-entropy loss. Extensive experiments on the Market-1501 and DukeMTMC-reID benchmarks demonstrate state-of-the-art performance of the proposed method. Code will be released to facilitate further studies on the unsupervised cross-domain person Re-ID task.

show abstract

Section: Market-1501mentioning

confidence: 99%

Domain Adaptive Attention Model for Unsupervised Cross-Domain Person Re-Identification

Huang,

Peng,

Jin

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…Second, we intend to study how to further exploit discriminative features from the last convolutional layers instead of using global average pooling. [38] 44.4 63.9 72.2 20.8 WARCA [13] 45.2 68.1 76.0 -KLFDA [14] 46.5 71.1 79.9 -SOMAnet [3] 73.9 --47.9 SVDNet [25] 82.3 92.3 95.2 62.1 PAN [40] 82.8 --63.4 Transfer [11] 83.7 --65.5 Triplet Loss [12] 84.9 94.2 -69.1 DML [34] 87.7 --68.8 MultiRegion [26] 66.4 85.0 90.2 41.2 HydraPlus [21] 76.9 91.3 94.5 -PAR [36] 81.0 92.0 94.7 -MultiLoss [16] 83.9 --64.4 PDC* [24] 84.4 92.7 94.9 63.4 PartLoss [31] 88.…”

Section: Discussionmentioning

confidence: 99%

Discovering Underlying Person Structure Pattern with Relative Local Distance for Person Re-identification

Wang¹,

Lai²,

Xie³

et al. 2019

Preprint

View full text Add to dashboard Cite

Modeling the underlying person structure for person reidentification (re-ID) is difficult due to diverse deformable poses, changeable camera views and imperfect person detectors. How to exploit underlying person structure information without extra annotations to improve the performance of person re-ID remains largely unexplored. To address this problem, we propose a novel Relative Local Distance (RLD) method that integrates a relative local distance constraint into convolutional neural networks (CNNs) in an end-to-end way. It is the first time that the relative local constraint is proposed to guide the global feature representation learning. Specially, a relative local distance matrix is computed by using feature maps and then regarded as a regularizer to guide CNNs to learn a structure-aware feature representation. With the discovered underlying person structure, the RLD method builds a bridge between the global and local feature representation and thus improves the capacity of feature representation for person re-ID. Furthermore, RLD also significantly accelerates deep network training compared with conventional methods. The experimental results show the effectiveness of RLD on the CUHK03, Market-1501, and DukeMTMC-reID datasets. Code is available at https://github.com/Wanggcong/RLD_codes.

show abstract

“…However, the Proposed model shows the best accuracy among the three. [22] 76.90 ---HP-Net [31] 76.90 ---PIE (poseBox) [5] 79.33 55.95 --DLPAR [32] 81.00 63.40 --SSM [33] 82.21 68.80 88.2 76.2 PDC [34] 84.14 63.41 --SVDNet (RE) [29] 87.08 71.31 --DML [35] 87.70 68.80 --DeepTransfer [36] 83.70 65.50 89.60 73.80 JLML [37] 85.10 65.50 89.70 73.80 MLFN [38] 90.00 74.30 --HA-CNN [39] 91.20 75.70 --PCB [14] 93.80 81.60 --US-GAN [11] 83.97 66.07 88.42 76.10 Liu et al [24] 87.65 68.92 --PN-GAN [26] 89 3) Robustness: Our pt-GAN model shows good robustness towards occlusion, illumination and scale. As seen in the person 1 of Figure 5, the occlusion (railings) present in the input image is not propagated in any of the generated images.…”

Section: B Pose Clusteringmentioning

confidence: 99%

Pose Invariant Person Re-Identification using Robust Pose-transformation GAN

Karmakar¹,

Mishra²

2021

Preprint

View full text Add to dashboard Cite

The objective of person re-identification (re-ID) is to retrieve a person's images from an image gallery, given a single instance of the person of interest. Despite several advancements, learning discriminative identity-sensitive and viewpoint invariant features for robust Person Re-identification is a major challenge owing to large pose variation of humans. This paper proposes a re-ID pipeline which utilizes the image generation capability of Generative Adversarial Networks combined with pose clustering and feature fusion to achieve pose invariant feature learning. The objective is to model a given person under different viewpoints and large pose changes and extract the most discriminative features from all the appearances. The pose transformational GAN (pt-GAN) module is trained to generate a person's image in any given pose. In order to identify the most significant poses for discriminative feature extraction, a Pose Clustering module is proposed. The given instance of the person is modelled in varying poses and these features are effectively combined through the Feature Fusion Network. The final re-ID model consisting of these 3 sub blocks, alleviates the pose dependence in person re-ID. Also, The proposed model is robust to occlusion, scale, rotation and illumination, providing a framework for viewpoint invariant feature learning. The proposed method outperforms the stateof-the-art GAN based models in 4 benchmark datasets. It also surpasses the state-of-the-art models that report higher re-ID accuracy in terms of improvement over baseline.The code snippets 1 of this work canbe found in https://github.com/arnabk001/Pose-Invariant-Person-Re-Identification-using-Robust-Pose-Transfromation-GAN

show abstract

Deep Transfer Learning for Person Re-identification

Cited by 99 publications

References 54 publications

Domain Adaptive Attention Model for Unsupervised Cross-Domain Person Re-Identification

Domain Adaptive Attention Model for Unsupervised Cross-Domain Person Re-Identification

Discovering Underlying Person Structure Pattern with Relative Local Distance for Person Re-identification

Pose Invariant Person Re-Identification using Robust Pose-transformation GAN

Contact Info

Product

Resources

About