融合注意力机制的师生网络无监督行人重识别

doi:10.19678/j.issn.1000-3428.0069486

摘要/Abstract

摘要：

无监督域自适应(UDA)行人重识别(Re-ID)技术致力于将已标记的源域知识转移到未标记的目标域, 但由于伪标签噪声和域间隙等问题的存在, 使得Re-ID十分具有挑战性。对此, 提出一种融合注意力机制的异构师生网络(HTSA)。该网络通过师生模型有效降低伪标签噪声的影响, 结合注意力机制关注行人关键信息, 滤除无关的背景信息; 采用域特定批处理归一化(DSBN)减弱由域间隙带来的性能下降问题; 采用一种新的数据增强方法, 将输入图像沿宽度方向切割为相同大小的两部分, 然后对每一部分进行独立的随机处理, 提升模型的泛化性能。实验结果显示, 在DukeMTMC-reID→MSMT17上, HTSA的均值平均精度(mAP)和Rank-1分别达到40.3%和71.0%, 在Market-1501→MSMT17上mAP和Rank-1分别达到37.7%和67.7%, 验证了HTSA的有效性。

关键词: 行人重识别, 无监督域自适应, 伪标签噪声, 注意力机制, 师生网络

Abstract:

Unsupervised Domain Adaptation (UDA) person Re-Identification (Re-ID) aims to transfer labeled source domain knowledge to an unlabeled target domain, which is very challenging owing to existing problems such as pseudo-label noise and domain gaps. Therefore, a Heterogeneous Teacher—Student network with Attention mechanisms (HTSA) is proposed to effectively reduce the influence of pseudo-label noise and focus on the key information of pedestrians while filtering out irrelevant background information. This study adopts Domain-Specific Batch Normalization (DSBN) to attenuate the performance degradation caused by domain gaps. Additionally, a novel data augmentation method is adopted to independently process two equally sized parts of the input image after width splitting, thereby enhancing the generalization ability. The experimental results reveal that mean Average Precision (mAP) and Rank-1 on DukeMTMC-reID→MSMT17 reach 40.3% and 71.0%, respectively, whereas those on Market-1501→MSMT17 reach 37.7% and 67.7%, respectively. This result demonstrates the effectiveness of the proposed method.

Key words: person Re-Identification (Re-ID), Unsupervised Domain Adaptation (UDA), pseudo-label noise, attention mechanism, teacher—student network

陈玉敏, 车进, 吴金蔓. 融合注意力机制的师生网络无监督行人重识别[J]. 计算机工程, 2025, 51(11): 304-316.

CHEN Yumin, CHE Jin, WU Jinman. Unsupervised Person Re-Identification for Teacher—Student Networks Incorporating Attention Mechanism[J]. Computer Engineering, 2025, 51(11): 304-316.

https://www.ecice06.com/CN/Y2025/V51/I11/304

图/表 21

图1 HTSA预训练网络结构

Fig.1 HTSA pre-training network structure

图2 HTSA微调网络结构

Fig.2 HTSA fine tunes network structure

图3 数据增强示意图

Fig.3 Schematic diagram of data augmentation

图4 背景噪声的影响

Fig.4 Impact of background noise

图5 CBAM结构

Fig.5 CBAM structure

图6 通道注意力模块

Fig.6 Channel attention module

图7 空间注意力模块

Fig.7 Spatial attention module

图8 非局部注意力模块

Fig.8 Non_local attention module

图9 BN和DSBN模块

Fig.9 BN and DSBN modules

图10 IBN-ResNet50残差块

Fig.10 IBN-ResNet50 residual block

图11 不同超参数下的HTSA网络性能评估

Fig.11 Performance evaluation of HTSA with different hyperparameters

图12 top-5检索结果可视化

Fig.12 Visualisation of top-5 retrieval results

图13 HTSA特征分布可视化

Fig.13 Visualisation of HTSA feature distribution

参考文献 39

1	蔡毅翔, 秦品乐, 曾建潮, 等. 针对大角度下视角差异的行人重识别方法研究. 计算机工程, 2024, 50 (5): 330- 341. doi: 10.19678/j.issn.1000-3428.0067532
	CAI Y X , QIN P L , ZENG J C , et al. Research on person re-identification method for large-angle viewpoint differences. Computer Engineering, 2024, 50 (5): 330- 341. doi: 10.19678/j.issn.1000-3428.0067532
2	SUN Y F, CHENG C M, ZHANG Y H, et al. Circle Loss: a unified perspective of pair similarity optimization[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2020: 6397-6406.
3	朱锦雷, 李艳凤, 陈后金, 等. 近邻优化跨域无监督行人重识别算法. 中国图象图形学报, 2023, 28 (11): 3471- 3484.
	ZHU J L , LI Y F , CHEN H J , et al. Cross-domain unsupervised Re-ID algorithm based on neighbor adversarial and consistency loss. Journal of Image and Graphics, 2023, 28 (11): 3471- 3484.
4	VERMA A , SUBRAMANYAM A V , WANG Z , et al. Unsupervised domain adaptation for person re-identification via individual-preserving and environmental-switching cyclic generation. IEEE Transactions on Multimedia, 2023, 25, 364- 377. doi: 10.1109/TMM.2021.3126404
5	GE Y X, CHEN D P, LI H S. Mutual mean-teaching: pseudo label refinery for unsupervised domain adaptation on person re-identification[EB/OL]. [2023-10-05]. https://arxiv.org/abs/2001.01526.
6	WANG W H , ZHAO F , LIAO S C , et al. Attentive WaveBlock: complementarity-enhanced mutual networks for unsupervised domain adaptation in person re-identification and beyond. IEEE Transactions on Image Processing, 2022, 31, 1532- 1544. doi: 10.1109/TIP.2022.3140614
7	ZHOU H , KONG J , JIANG M , et al. Heterogeneous dual network with feature consistency for domain adaptation person re-identification. International Journal of Machine Learning and Cybernetics, 2023, 14 (5): 1951- 1965. doi: 10.1007/s13042-022-01739-9
8	ZHONG X , HAN X Y , JIA X M , et al. ICLR: instance credibility-based label refinement for label noisy person re-identification. Pattern Recognition, 2024, 148, 110168. doi: 10.1016/j.patcog.2023.110168
9	余志敏. 复杂背景下的跨域行人重识别技术研究[D]. 长沙: 中南大学, 2022.
	YU Z M. Research on cross-domain person re-identification with complex background[D]. Changsha: Central South University, 2022. (in Chinese)
10	ISOBE T, LI D, TIAN L, et al. Towards discriminative representation learning for unsupervised person re-identification[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Washington D.C., USA: IEEE Press, 2021: 8506-8516.
11	LUO C C , SONG C F , ZHANG Z X . Learning to adapt across dual discrepancy for cross-domain person re-identification. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45 (2): 1963- 1980. doi: 10.1109/TPAMI.2022.3167053
12	CHEN H, LAGADEC B, BREMOND F. Enhancing diversity in teacher-student networks via asymmetric branches for unsupervised person re-identification[C]//Proceedings of the IEEE Winter Conference on Applications of Computer Vision (WACV). Washington D.C., USA: IEEE Press, 2021: 1-10.
13	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[EB/OL]. [2023-10-05]. https://arxiv.org/abs/1807.06521.
14	HAN J L, FANG P F, LI W H, et al. You Only Cut Once: boosting data augmentation with a single cut[EB/OL]. [2023-10-05]. https://arxiv.org/pdf/2201.12078.
15	WANG X L, GIRSHICK R, GUPTA A, et al. Non-local neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D.C., USA: IEEE Press, 2018: 7794-7803.
16	LI K Q , WU L , QI Q , et al. Beyond single reference for training: underwater image enhancement via comparative learning. IEEE Transactions on Circuits and Systems for Video Technology, 2023, 33 (6): 2561- 2576. doi: 10.1109/TCSVT.2022.3225376
17	MA X , LV W Q , ZHAO M . A double stream person re-identification method based on attention mechanism and multi-scale feature fusion. IEEE Access, 2023, 11, 14612- 14620. doi: 10.1109/ACCESS.2023.3243553
18	IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift[EB/OL]. [2023-10-05]. https://arxiv.org/abs/1502.03167.
19	CHANG W G, YOU T, SEO S, et al. Domain-specific batch normalization for unsupervised domain adaptation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2019: 7346-7354.
20	SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2016: 2818-2826.
21	HERMANS A, BEYER L, LEIBE B. In defense of the triplet loss for person re-identification[EB/OL]. [2023-10-05]. https://arxiv.org/abs/1703.07737.
22	ZHENG L, SHEN L Y, TIAN L, et al. Scalable person re-identification: a benchmark[C]//Proceedings of the IEEE International Conference on Computer Vision (ICCV). Washington D.C., USA: IEEE Press, 2015: 1116-1124.
23	ZHENG Z D, ZHENG L, YANG Y. Unlabeled samples generated by GAN improve the person re-identification baseline in vitro[C]//Proceedings of the IEEE International Conference on Computer Vision (ICCV). Washington D.C., USA: IEEE Press, 2017: 3774-3782.
24	WEI L H, ZHANG S L, GAO W, et al. Person transfer GAN to bridge domain gap for person re-identification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D.C., USA: IEEE Press, 2018: 79-88.
25	ZHONG Z, ZHENG L, CAO D L, et al. Re-ranking person re-identification with k-reciprocal encoding[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2017: 3652-3661.
26	LI S H , YUAN M K , CHEN J , et al. AdaDC: adaptive deep clustering for unsupervised domain adaptation in person re-identification. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32 (6): 3825- 3838. doi: 10.1109/TCSVT.2021.3118060
27	GU J Y, WANG K, LUO H, et al. MSINet: twins contrastive search of multi-scale interaction for object ReID[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2023: 19243-19253.
28	FU Y, WEI Y C, WANG G S, et al. Self-similarity grouping: a simple unsupervised cross domain adaptation approach for person re-identification[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Washington D.C., USA: IEEE Press, 2019: 6111-6120.
29	ZHAO F, LIAO S C, XIE G S, et al. Unsupervised domain adaptation with noise resistible mutual-training for person re-identification[EB/OL]. [2023-10-05]. https://link.springer.com/chapter/10.1007/978-3-030-58621-8_31.
30	ZHENG K C , LAN C L , ZENG W J , et al. Exploiting sample uncertainty for domain adaptive person re-identification. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35 (4): 3538- 3546. doi: 10.1609/aaai.v35i4.16468
31	GE Y X, ZHU F, CHEN D P, et al. Self-paced contrastive learning with hybrid memory for domain adaptive object Re-ID[EB/OL]. [2023-10-05]. https://arxiv.org/abs/2006.02713.
32	DAI Y X , LIU J , BAI Y , et al. Dual-refinement: joint label and feature refinement for unsupervised domain adaptive person re-identification. IEEE Transactions on Image Processing, 2021, 30, 7815- 7829. doi: 10.1109/TIP.2021.3104169
33	ZHENG K C, LIU W, HE L X, et al. Group-aware label transfer for domain adaptive person re-identification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2021: 5306-5315.
34	HAN J , LI Y L , WANG S J . Delving into probabilistic uncertainty for unsupervised domain adaptive person re-identification. Proceedings of the AAAI Conference on Artificial Intelligence, 2022, 36 (1): 790- 798. doi: 10.1609/aaai.v36i1.19960
35	HE T , SHEN L Q , GUO Y C , et al. SECRET: self-consistent pseudo label refinement for unsupervised domain adaptive person re-identification. Proceedings of the AAAI Conference on Artificial Intelligence, 2022, 36 (1): 879- 887. doi: 10.1609/aaai.v36i1.19970
36	DONG W H , QU P S , LI B W . Dual pseudo label refinement for unsupervised domain adaptive person re-identification. IEEE Access, 2023, 11, 44402- 44412. doi: 10.1109/ACCESS.2023.3272879
37	YANG Q Z, YU H X, WU A C, et al. Patch-based discriminative feature learning for unsupervised person re-identification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D.C., USA: IEEE Press, 2019: 3628-3637.
38	ZOU Y, YANG X D, YU Z D, et al. Joint disentangling and adaptation for cross-domain person re-identification[EB/OL].[2023-10-05]. https://arxiv.org/abs/2007.10315.
39	LAURENS V D M, HINTON G. Visualizing data using t-SNE[EB/OL].[2023-10-05]. https://jmlr.org/papers/v9/vandermaaten08a.html.

[1]	黄金贵, 刘朋, 唐文胜. MMD-YOLOv7:黑暗条件下车辆检测方法[J]. 计算机工程, 2025, 51(9): 340-349.
[2]	符家成, 田瑾, 张玉金, 方志军. 结合前置三元组集的知识图谱推荐[J]. 计算机工程, 2025, 51(9): 101-109.
[3]	翟志鹏, 曹阳, 沈琴琴, 施佺. 基于多时空图融合与动态注意力的交通流预测[J]. 计算机工程, 2025, 51(9): 139-148.
[4]	马淦, 谷雨, 彭冬亮. 结合改进YOLOv5s和动态数据增强的海面舰船检测[J]. 计算机工程, 2025, 51(9): 294-305.
[5]	陈彦如, 刘珂良, 冉茂亮. 基于深度强化学习的外卖即时配送实时优化[J]. 计算机工程, 2025, 51(9): 328-339.
[6]	张昭理, 李家豪, 刘海, 石佛波, 何嘉文. 基于个性化遗忘建模的知识追踪方法[J]. 计算机工程, 2025, 51(8): 120-130.
[7]	闫建红, 刘芝妍, 王震. 融合时空注意力机制的多尺度卷积车辆轨迹预测[J]. 计算机工程, 2025, 51(8): 406-414.
[8]	倪源松, 韩军, 邹小燕, 胡广怡, 王文帅. 两阶段自适应分块输电线路螺栓缺陷检测方法[J]. 计算机工程, 2025, 51(8): 281-291.
[9]	郝宏达, 罗健旭. 基于多尺度区域特征融合的多器官语义分割模型[J]. 计算机工程, 2025, 51(8): 270-280.
[10]	宋杰, 徐慧英, 朱信忠, 黄晓, 陈晨, 王泽宇. 基于YOLOv8改进的跌倒检测算法: OEF-YOLO[J]. 计算机工程, 2025, 51(7): 127-139.
[11]	彭菊红, 张弛, 高谦, 张光明, 谈栋华, 赵明俊. 基于改进的YOLOv8算法的钢材缺陷检测[J]. 计算机工程, 2025, 51(7): 152-160.
[12]	刘春霞, 孟吉星, 潘理虎, 龚大立. 融合RGB与IR图像的遥感小目标检测方法[J]. 计算机工程, 2025, 51(7): 326-338.
[13]	栾孟娜, 郑秋梅, 王风华. 基于DMC-YOLO的交通标志实时检测算法[J]. 计算机工程, 2025, 51(7): 90-99.
[14]	田青, 王斌, 周子枭. 无监督行人重识别研究综述[J]. 计算机工程, 2025, 51(7): 12-30.
[15]	华家宝, 张京瑞, 朱福民, 陈璐. 基于路侧相机的自适应空间变换车辆检测方法[J]. 计算机工程, 2025, 51(6): 349-359.

选择文件类型/文献管理软件名称

选择包含的内容