

浏览全部资源
扫码关注微信
中山大学电子与信息工程学院, 广东 广州 510006
Received:12 April 2025,
Revised:2025-05-13,
Accepted:07 May 2025,
Published Online:04 July 2025,
Published:25 September 2025
移动端阅览
郑昊天,胡海峰.知识蒸馏与掩码重构的域泛化行人重识别[J].中山大学学报(自然科学版)(中英文),2025,64(05):43-49.
ZHENG Haotian,HU Haifeng.A distillation and masked approach for domain generalizable person re-identification[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2025,64(05):43-49.
郑昊天,胡海峰.知识蒸馏与掩码重构的域泛化行人重识别[J].中山大学学报(自然科学版)(中英文),2025,64(05):43-49. DOI: 10.13471/j.cnki.acta.snus.ZR20250070.
ZHENG Haotian,HU Haifeng.A distillation and masked approach for domain generalizable person re-identification[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2025,64(05):43-49. DOI: 10.13471/j.cnki.acta.snus.ZR20250070.
域泛化行人重识别的挑战源于当前基准方法的2个固有局限性:1)数据集之间存在明显的域间隙,2)数据集域内多样性不足。现有一些多领域联合训练方法,往往无法充分学习跨域数据集间潜在的身份线索。为了克服上述局限,本文通过一种双分支策略来增强模型泛化性能。首先针对大规模预训练的扩展模型进行知识蒸馏,同时针对现有多域训练数据进行掩码图像特征挖掘。常用的域泛化行人重识别协议基准上的实验证明了本文方法的性能。在以Market-1501为目标域的留一法测试中,本文方法相对于基准方法提高了16.2%的Rank-1准确度,相对现存最佳方法则在Rank-1准确度上实现了3.6%的提升。
The challenge of domain generalization stems from two inherent limitations in current person re-identification benchmarks:1)significant inter-dataset domain gaps, and 2) insufficient intra-dataset diversity. While existing multi-domain joint training approaches attempt to address these issues, they often fail to fully exploit latent discriminative identity cues across datasets. To address the aforementioned limitations,our framework enhances network generalization capabilities through a dual-branch strategy: knowledge distillation employed from a large-scale pre-trained model along with mask image feature mining performed on existing multi-domain training data. Extensive experiments on popular domain generalization person ReID benchmarks demonstrate that our method can achieve superior performance. Notably, our approach achieves a 16.2% Rank-1 accuracy gain over the baseline and a 3.6% improvement over existing state-of-the-art methods under the leave-one-out protocol using Market-1501.
冯展祥 , 朱荣 , 王玉娟 , 等 , 2020 . 非可控环境行人再识别综述 [J]. 中山大学学报(自然科学版中英文) , 59 ( 3 ): 1 - 11 .
郭迎春 , 冯放 , 阎刚 , 等 , 2022 . 基于自适应融合网络的跨域行人重识别方法 [J]. 自动化学报 , 48 ( 11 ): 2744 - 2756 .
叶钰 , 王正 , 梁超 , 等 , 2020 . 多源数据行人重识别研究综述 [J]. 自动化学报 , 46 ( 9 ): 01869 - 01884 .
朱锦雷 , 李艳凤 , 陈后金 , 等 , 2023 . 近邻优化跨域无监督行人重识别算法 [J]. 中国图象图形学报 , 28 ( 11 ): 3471 - 3484 .
DAI Y , LI X , LIU J , et al , 2021 . Generalizable person re-identification with relevance-aware mixture of experts [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . Nashville ,TN,USA : 16140 - 16149 .
DEVLIN J , CHANG M W , LEE K , et al , 2019 . Bert: Pre-training of deep bidirectional transformers for language understanding [C]// The North American Chapter of the Association for Computational Linguistics: Human Language Technologies . Minneapolis,MN,USA : 4171 - 4186 .
DOSOVITSKIY A , BEYER L , KOLESNIKOV A , et al , 2020 . An image is worth 16 x 16 words: Transformers for image recognition at scale[EB/OL]. arXiv: 2010.11929 .
DOU Z , WANG Z , LI Y , et al , 2023 . Identity-seeking self-supervised representation learning for generalizable person re-identification [C]// IEEE/CVF International Conference on Computer Vision . Paris,France : 15801 - 15812 .
ERGASTI A , FONTANINI T , FERRARI C , et al , 2024 . Mars: Paying more attention to visual attributes for text-based person search [EB/OL]. arXiv . 2407 . 04287 .
FU D , CHEN D , BAO J , et al , 2021 . Unsupervised pre-training for person re-identification [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . Nashville, TN, USA : 14745 - 14754 .
HE L , LIU W , LIANG J , et al , 2021a . Semi-supervised domain generalizable person re-identification [EB/OL]. arXiv: 2108.05045 .
HE S , LUO H , WANG P , et al , 2021 b. Transreid: Transformer-based object re-identification[C]// IEEE/CVF International Conference on Computer Vision . Montreal, QC, Canada : 14993 - 15002 .
LI W , ZHAO R , XIAO T , et al , 2014 . DeepReID: Deep filter pairing neural network for person re-identification [C]// IEEE Conference on Computer Vision and Pattern Recognition . Columbus, OH, USA : 152 - 159 .
LV K , CHEN H , ZHAO C , et al , 2024 . Style variable and irrelevant learning for generalizable person re-identification [J]. ACM Trans Multimedia Comput Commun Appl , 20 ( 9 ): 1 - 22 .
MA H , LI X , YUAN X , et al , 2023 . Two-phase self-supervised pretraining for object re-identification [J]. Knowl Based Syst , 261 : 0110220 .
RUSSAKOVSKY O , DENG J , SU H , et al , 2015 . ImageNet large scale visual recognition challenge [J]. Int J Comput Vis , 115 ( 3 ): 211 - 252 .
WANG F , LIU H , 2021 . Understanding the behaviour of contrastive loss [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . Nashville, TN, USA : 2495 - 2504 .
WEI L , ZHANG S , GAO W , et al , 2018 . Person transfer gan to bridge domain gap for person re-identification [C]// IEEE Conference on Computer Vision and Pattern Recognition . Salt Lake City, UT, USA : 79 - 88 .
YANG S , ZHOU Y , ZHENG Z , et al , 2023 . Towards unified text-based person retrieval: A large-scale multi-attribute and language search benchmark [C]// 31st ACM International Conference on Multimedia . Ottawa,ON,Canada : 4492 - 4501 .
ZHAO Y , WANG G , LUO C , et al , 2021 . Self-supervised visual representations learning by contrastive mask prediction [C]// IEEE/CVF International Conference on Computer Vision . Montreal, QC, Canada : 10160 - 10169 .
ZHENG L , SHEN L , TIAN L , et al , 2015 . Scalable person re-identification:A benchmark [C]// IEEE International Conference on Computer Vision . Santiago, Chile : 1116 - 1124 .
ZHENG Z , ZHENG L , YANG Y , 2017 . Unlabeled samples generated by gan improve the person re-identification baseline in vitro [C]// IEEE International Conference on Computer Vision . Venice, Italy : 3754 - 3762 .
0
Views
126
下载量
0
CSCD
Publicity Resources
Related Articles
Related Author
Related Institution
京公网安备11010802024621