面向跨区域场景的无监督域自适应行人重识别

doi:10.16183/j.cnki.jsjtu.2023.635

摘要/Abstract

摘要：

在大规模监控系统中,由于跨区域场景间的距离较远,从不同区域的相机中获取行人正样本变得极为困难,这限制了行人重识别模型在跨区域场景中的有效应用.为解决跨区域场景中跨相机缺乏正样本的问题,提出一种多粒度特征挖掘和域不变特征学习的无监督域自适应行人重识别方法.该方法主要包含多粒度特征学习模块和域分布对齐模块.在多粒度特征学习模块中,通过全局特征学习提取行人的全局判别性特征.为进一步提升所提取行人特征的判别性,提出了局部一致性特征学习模块来加强行人局部特征之间的交互.通过全局和局部特征的学习,促进网络提取行人多粒度的判别性特征来提升行人重识别模型的性能.此外,设计了域分布对齐模块,通过风格迁移为目标域数据样本构建跨相机不同风格的正样本,解决了跨区域场景中跨相机缺乏正样本的问题,同时提升了模型的域自适应能力.在Market-1501、DukeMTMC、CUHK03和MSMT17数据集上的实验表明,所提方法相较于当前先进的域自适应行人重识别方法具有明显优势.

关键词: 行人重识别, 域自适应, 多粒度特征挖掘, 域分布对齐

Abstract:

In large-scale surveillance systems, the lack of positive cross-camera pedestrian samples in cross-regional scenes limits the performance of person re-identification (Re-ID) models. To tackle this challenge, an unsupervised domain adaptive person re-identification method incorporating multi-granularity feature mining and domain-invariant feature learning is proposed. The method consists of a multi-granularity feature learning module and a domain distribution alignment module. Within the multi-granularity feature learning module, global discriminant features of pedestrians are extracted through global features learning. To further enhance the discriminative capability of pedestrian features, a local consistency feature learning module is proposed to strengthen the interactions among local features. By jointly learning global and local features, the network is encouraged to extract multi-granularity discriminative features, thereby improving the performance of the person re-identification model. Additionally, a domain distribution alignment module is incorporated, leveraging style transfer to generate positive samples with diverse styles across cameras for target domain. This not only addresses the issue of the lack of positive samples across cameras in cross-regional scenes but also enhances the domain adaptation capabilities of the model. Extensive experiments conducted on the Market-1501, DukeMTMC, CUHK03, and MSMT17 datasets demonstrate the effectiveness of the proposed method compared to state-of-the-art person re-identification methods.

Key words: person re-identification (Re-ID), domain adaptation, multi-granularity feature mining, domain distribution alignment

中图分类号:

TP391.41

毛彦嵋, 李华锋, 张亚飞. 面向跨区域场景的无监督域自适应行人重识别[J]. 上海交通大学学报, 2025, 59(12): 1878-1890.

MAO Yanmei, LI Huafeng, ZHANG Yafei. Unsupervised Domain Adaptation for Cross-Regional Scenes Person Re-Identification[J]. Journal of Shanghai Jiao Tong University, 2025, 59(12): 1878-1890.

图/表 7

图1

表1

图2

图3

表2

表3

表4

参考文献 47

[1]	GONG S, CRISTANI M, LOY C C, et al. The re-identification challenge[M]. London,UK: Springer, 2014.
[2]	李擎, 皇甫玉彬, 李江昀, 等. UConvTrans: 全局和局部信息交互的双分支心脏图像分割[J]. 上海交通大学学报, 2023, 57(5): 570-581. doi: 10.16183/j.cnki.jsjtu.2022.088
	LI Qing, HUANGFU Yubin, LI Jiangyun, et al. UConvTrans: Dual-branch cardiac image segmentation with global and local information interaction[J]. Journal of Shanghai Jiao Tong University, 2023, 57(5): 570-581.
[3]	高涛, 文渊博, 陈婷, 等. 基于窗口自注意力网络的单图像去雨算法[J]. 上海交通大学学报, 2023, 57(5): 613-623. doi: 10.16183/j.cnki.jsjtu.2022.032
	GAO Tao, WEN Yuanbo, CHEN Ting, et al. Single image deraining algorithm based on window self-attention network[J]. Journal of Shanghai Jiao Tong University, 2023, 57(5): 613-623.
[4]	万安平, 杨洁, 缪徐, 等. 基于注意力机制与神经网络的热电联产锅炉负荷预测[J]. 上海交通大学学报, 2023, 57(3): 316-325. doi: 10.16183/j.cnki.jsjtu.2021.346
	WAN Anping, YANG Jie, MIAO Xu, et al. Load prediction of combined heat and power boilers based on attention mechanism and neural network[J]. Journal of Shanghai Jiao Tong University, 2023, 57(3): 316-325.
[5]	石林波, 李华锋, 张亚飞, 等. 模态不变性特征学习和一致性细粒度信息挖掘的跨模态行人重识别[J]. 模式识别与人工智能, 2022, 35(12): 1064-1077. doi: 10.16451/j.cnki.issn1003-6059.202212002
	SHI Linbo, LI Huafeng, ZHANG Yafei, et al. Cross-modal person re-identification with modality-invariant feature learning and consistent fine-grained information mining[J]. Pattern Recognition and Artificial Intelligence, 2022, 35(12): 1064-1077. doi: 10.16451/j.cnki.issn1003-6059.202212002
[6]	ZHONG Z, ZHENG L, LUO Z, et al. Invariance matters: Exemplar memory for domain adaptive person re-identification[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Los Angeles, USA: IEEE, 2019: 598-607.
[7]	YU H X, ZHENG W S, WU A, et al. Unsupervised person re-identification by soft multilabel learning[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Los Angeles, USA: IEEE, 2019: 2148-2157.
[8]	LIU J, ZHA Z J, CHEN D, et al. Adaptive transfer network for cross-domain person re-identification[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Los Angeles, USA: IEEE, 2019: 7202-7211.
[9]	李玲莉, 张亚飞, 李华锋, 等. 低秩先验引导的无监督域自适应行人重识别[J]. 重庆大学学报, 2021, 44(11): 57-70.
	LI Lingli, ZHANG Yafei, LI Huafeng, et al. Unsupervised domain adaptive person re-identification guided by low-rank prior[J]. Journal of Chongqing University, 2021, 44(11): 57-70.
[10]	GE Y, CHEN D, LI H. Mutual mean-teaching: Pseudo label refinery for unsupervised domain adaptation on person re-identification[DB/OL]. (2020-06-01)[2023-12-19]. https://arxiv.org/abs/2001.01526.
[11]	YANG F, LI K, ZHONG Z, et al. Asymmetric co-teaching for unsupervised cross-domain person re-identification[C]// Proceedings of the AAAI Conference on Artificial Intelligence (AAAI). New York, USA: AAAI, 2020, 34(7): 12597-12604.
[12]	LI Z, SHI Y, LING H, et al. Reliability exploration with self-ensemble learning for domain adaptive person re-identification[C]// Proceedings of the AAAI conference on Artificial Intelligence (AAAI). Vancouver, Canada: AAAI, 2022, 36(2): 1527-1535.
[13]	ZHENG D, XIAO J, CHEN K, et al. Soft pseudo-label shrinkage for unsupervised domain adaptive person re-identification[J]. Pattern Recognition, 2022, 127: 108615. doi: 10.1016/j.patcog.2022.108615 URL
[14]	TANG C, XUE D, CHEN D. Feature diversity learning with sample dropout for unsupervised domain adaptive person re-identification[J]. Multimedia Tools and Applications, 2024, 83(2): 5079-5097. doi: 10.1007/s11042-023-15546-z
[15]	LEE G, LEE S, KIM D, et al. Camera-driven representation learning for unsupervised domain adaptive person re-identification[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Paris, France: IEEE, 2023: 11453-11462.
[16]	CHEN S, QIU L, TIAN Z, et al. MTNet: Mutual tri-training network for unsupervised domain adaptation on person re-identification[J]. Journal of Visual Communication and Image Representation, 2023, 90: 103749. doi: 10.1016/j.jvcir.2022.103749 URL
[17]	LI S, LI F, LI J, et al. Logical relation inference and multiview information interaction for domain adaptation person re-identification[J]. IEEE Transactions on Neural Networks and Learning Systems (TNNLS), 2023, 34(10): 14770-14782.
[18]	GE Y, ZHU F, CHEN D, et al. Structured domain adaptation with online relation regularization for unsupervised person Re-ID[J]. IEEE Transactions on Neural Networks and Learning Systems (TNNLS), 2024, 35(1): 258-271.
[19]	SU C, LI J, GAO W, et al. Pose-driven deep convolutional model for person re-identification[C]// Proceedings of the IEEE International Conference on Computer Vision (ICCV). Venice, Italy: IEEE, 2017, 3960-3969.
[20]	ZHONG Z, ZHENG L, ZHENG Z, et al. Camera style adaptation for person re-identification[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Salt Lake City, USA: IEEE, 2018: 5157-5166.
[21]	YANG F, YAN K, LU S, et al. Part-aware progressive unsupervised domain adaptation for person re-identification[J]. IEEE Transactions on Multi-media, 2020, 23: 1681-1695.
[22]	ULYANOV D, VEDALDI A, LEMPITSKY V. Instance normalization: The missing ingredient for fast stylization[DB/OL]. (2017-11-06)[2023-12-19]. https://arxiv.org/abs/1607.08022.
[23]	DUMOULIN V, SHLENS J, KUDLUR M. A learned representation for artistic style[DB/OL]. (2017-02-09)[2023-12-19]. https://arxiv.org/abs/1610.07629.
[24]	DAI Y, SUN Y, LIU J, et al. Bridging the source-to-target gap for cross-domain person re-identification with intermediate domains[J]. International Journal of Computer Vision, 2025, 133(1): 410-434. doi: 10.1007/s11263-024-02169-6
[25]	HUANG X, BELONGIE S. Arbitrary style transfer in real-time with adaptive instance normalization[C]// Proceedings of the IEEE International Conference on Computer Vision (ICCV). Venice, Italy: IEEE, 2017: 1501-1510.
[26]	LI H, MAO Y, ZHANG Y, et al. Domain-adaptive person re-identification without cross-camera paired samples[J]. Engineering Applications of Artificial Intelligence, 2025, 145: 110171. doi: 10.1016/j.engappai.2025.110171 URL
[27]	ZHANG T, XIE L, WEI L, et al. Single camera training for person re-identification[C]// Proceedings of the AAAI Conference on Artificial Intelligence(AAAI). New York, USA: AAAI, 2020, 34(7): 12878-12885.
[28]	GE W, PAN C, WU A, et al. Cross-camera feature prediction for intra-camera supervised person re-identification across distant scenes[C]// Proceedings of the 29th ACM International Conference on Multimedia. Chengdu, China: ACM, 2021: 3644-3653.
[29]	WU C, GE W, WU A, et al. Camera-conditioned stable feature generation for isolated camera supervised person re-identification[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR). New Orleans, USA: IEEE, 2022: 20238-20248.
[30]	HE S, LUO H, WANG P, et al. Transreid: Transformer-based object re-identification[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision(ICCV). Virtual: IEEE, 2021: 15013-15022.
[31]	NI H, LI Y, GAO L, et al. Part-aware transformer for generalizable person re-identification[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Paris, France: IEEE, 2023: 11280-11289.
[32]	ZHENG L, SHEN L, TIAN L, et al. Scalable person re-identification: A benchmark[C]// Proceedings of the IEEE International Conference on Computer Vision (ICCV). Santiago, Chile: IEEE, 2015: 1116-1124.
[33]	RISTANI E, SOLERA F, ZOU R, et al. Performance measures and a data set for multi-target, multi-camera tracking[C]// European Conference on Computer Vision (ECCV). Amsterdam, Netherlands: UvA, 2016: 17-35.
[34]	WEI L, ZHANG S, GAO W, et al. Person transfer gan to bridge domain gap for person re-identification[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Salt Lake, USA: IEEE, 2018: 79-88.
[35]	LI W, ZHAO R, XIAO T, et al. Deepreid: Deep filter pairing neural network for person re-identification[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Columbus, USA: IEEE, 2014: 152-159.
[36]	WANG X, DORETTO G, SEBASTIAN T, et al. Shape and appearance context modeling[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Minnesota, USA: IEEE, 2007: 1-8.
[37]	ROBBINS H, MONRO S. A stochastic approximation method[J]. Annals of Mathematical Statistics, 1951, 22(3): 400-407. doi: 10.1214/aoms/1177729586 URL
[38]	PASZKE A, GROSS S, MASSA F, et al. PyTorch:An imperative style, high-performance deep learning library[C]// Advances in Neural Information Processing Systems (NeurIPS).Vancouver, Canada: Morgan Kaufmann,2019:8024-8035.
[39]	LUO H, GU Y, LIAO X, et al. Bag of tricks and a strong baseline for deep person re-identification[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Long Beach, USA: IEEE, 2019: 1487-1495.
[40]	GE Y, ZHU F, CHEN D, et al. Self-paced contrastive learning with hybrid memory for domain adaptive object re-id[J]. Advances in Neural Information Processing Systems, 2020, 33: 11309-11321.
[41]	ZHAI Y, YE Q, LU S, et al. Multiple expert brainstorming for domain adaptive person re-identification[C]// Proceedings of the European Conference on Computer Vision (ECCV). Glasgow, UK: EACV, 2020: 594-611.
[42]	LI H, PANG J, TAO D, et al. Cross adversarial consistency self-prediction learning for unsupervised domain adaptation person re-identification[J]. Information Sciences, 2021, 559: 46-60. doi: 10.1016/j.ins.2021.01.016 URL
[43]	DAI Y, LIU J, SUN Y, et al. IDM: An intermediate domain module for domain adaptive person re-ID[C]// Proceedings of the IEEE International Conference on Computer Vision (ICCV). Virtual: IEEE, 2021: 11864-11874.
[44]	DAI Y, LI J, BAI Y, et al. Dual-refinement: Joint label and feature refinement for unsupervised domain adaptive person re-identification[J]. IEEE Transactions on Image Processing (TIP), 2021, 30: 7815-7829.
[45]	HAN J, LI Y L, WANG S. Delving into probabilistic uncertainty for unsupervised domain adaptive person re-identification[C]// Proceedings of the AAAI Conference on Artificial Intelligence (AAAI). Vancouver, Canada: AAAI, 2022, 36(1): 790-798.
[46]	HE T, SHEN L, GUO Y, et al. SECRET: Self-consistent pseudo label refinement for unsupervised domain adaptive person re-identification[C]// Proceedings of the AAAI Conference on Artificial Intelligence (AAAI). Vancouver, Canada: AAAI, 2022, 36(1): 879-887.
[47]	LI H, XU K, LI J, et al. Dual-stream reciprocal disentanglement learning for domain adaptation person re-identification[J]. Knowledge-Based Systems (KBS), 2022, 251: 109315.

方法	MSMT→Market-SCT				Market→MSMT-SCT
方法	R1	R5	R10	mAP	R1	R5	R10	mAP
Baseline	71.5	83.9	87.9	49.5	38.6	51.5	57.3	16.7
Baseline + MFLM	77.6	89.1	93.0	57.3	42.4	55.2	61.3	21.7
Baseline + MFLM + DDAM	80.6	91.0	94.2	60.9	43.6	56.3	62.7	22.9

方法	MSMT17→Market-SCT				CUHK03→Market-SCT
方法	R1	R5	R10	mAP	R1	R5	R10	mAP
MMT-500^[10]	43.0	59.1	66.0	23.3	41.6	57.5	65.6	22.6
MMT-700^[10]	42.9	59.9	67.3	23.5	42.3	58.2	66.4	22.8
MMT-900^[10]	42.1	58.7	66.5	22.5	43.5	59.5	66.8	23.4
SPCL^[40]	15.1	25.2	32.3	6.7	13.5	23.8	30.4	5.8
MEB-Net^[41]	62.2	76.4	81.6	34.2	47.3	66.7	74.0	26.6
CAC^[42]	72.1	84.1	88.7	40.4	69.8	82.3	86.1	42.9
IDM^[43]	22.7	30.5	35.3	13.6	21.3	28.9	33.1	12.5
Dual-Refine^[44]	50.3	65.7	72.4	28.3	46.6	61.0	67.8	25.2
P2LR^[45]	44.3	59.2	65.9	25.3	$\underset{\_}{77.0}$	$\underset{\_}{90.4}$	$\underset{\_}{94.2}$	$\underset{\_}{57.3}$
SPLR^[13]	61.5	77.7	82.2	33.6	69.2	83.5	91.6	41.5
SECRET^[46]	57.5	74.1	80.0	31.1	57.8	73.5	78.5	32.8
DRDL^[47]	$\underset{\_}{77.4}$	$\underset{\_}{89.6}$	$\underset{\_}{93.1}$	$\underset{\_}{45.6}$	75.9	87.8	91.2	46.6
LRIMV^[17]	54.8	70.0	77.0	29.1	60.1	76.2	82.1	36.0
本文方法	80.6	91.0	94.2	60.9	82.5	93.0	95.9	65.3

方法	Duke→Market-SCT				Market→Duke-SCT
方法	R1	R5	R10	mAP	R1	R5	R10	mAP
MMT-500^[10]	50.0	68.0	75.9	27.8	38.9	56.3	63.5	26.8
MMT-700^[10]	49.1	66.9	74.3	27.7	40.9	58.1	65.5	29.2
MMT-900^[10]	51.0	70.0	76.9	28.5	42.3	59.6	67.6	30.4
SPCL^[40]	11.5	23.5	30.2	4.5	12.3	19.7	24.2	5.6
MEB-Net^[41]	54.4	71.1	78.1	30. 7	41.6	58.1	64.0	27.8
CAC^[42]	62.1	76.6	81.1	30.6	49.6	64.0	69.8	30.0
IDM^[43]	32.3	48.3	56.1	14.3	37.9	51.2	58.4	23.6
Dual-Refine^[44]	47.7	63.4	70.1	23.3	39.8	53.4	60.2	28.1
P2LR^[45]	52.6	68.5	75.1	25.9	35.7	49.8	56.4	20.6
SPLR^[13]	60.2	74.8	79.8	31.3	47.4	62.0	67.7	30.4
SECRET^[46]	56.5	71.0	77.2	28.5	43.0	58.4	64.5	27.6
DRDL^[47]	60.8	76.6	81.2	27.7	63.4	75.1	78.3	41.6
LRIMV^[17]	$\underset{\_}{63.1}$	$\underset{\_}{78.3}$	$\underset{\_}{82.7}$	$\underset{\_}{32.5}$	$\underset{\_}{64.1}$	$\underset{\_}{77.6}$	$\underset{\_}{79.1}$	$\underset{\_}{43.7}$
本文方法	71.0	83.6	87.9	46.7	67.2	78.9	82.5	48.2

方法	Market→MSMT-SCT				CUHK03→MSMT-SCT
方法	R1	R5	R10	mAP	R1	R5	R10	mAP
MMT-1000^[10]	15.9	25.7	31.0	6.4	30.6	44.1	50.8	13.3
MMT-2000^[10]	16.4	25.9	31.6	6.5	30.9	44.4	51.1	13.5
SPCL^[40]	7.1	9.7	12.9	3.9	13.7	22.5	28.6	5.2
MEB-Net^[41]	15.9	24.2	30.1	5.6	28.9	42.3	49.6	12.8
CAC^[42]	31.6	43.3	48.6	11.9	26.8	38.6	44.5	9.6
IDM^[43]	12.6	20.2	24.9	5.8	25.2	37.6	43.9	12.3
Dual-Refine^[44]	17.5	28.0	32.9	6.4	14.9	24.4	30.1	5.6
P2LR^[45]	15.5	24.3	29.4	6.2	15.1	24.0	29.0	5.9
SPLR^[13]	19.6	28.6	33.8	7.2	19.2	28.4	33.2	6.8
SECRET^[46]	18.8	28.2	33.2	6.8	17.1	26.2	31.5	5.6
DRDL^[47]	$\underset{\_}{37.6}$	$\underset{\_}{49.9}$	$\underset{\_}{55.5}$	$\underset{\_}{14.7}$	$\underset{\_}{35.8}$	$\underset{\_}{48.0}$	$\underset{\_}{53.5}$	$\underset{\_}{13.9}$
LRIMV^[17]	19.2	30.0	35.6	6.6	23.1	35.7	41.7	8.2
本文方法	43.6	56.3	62.7	22.9	42.5	55.7	62.4	22.2