基于GAN的无监督域自适应行人重识别

doi:10.11959/j.issn.1000-0801.2021016

电信科学 ›› 2021, Vol. 37 ›› Issue (2): 99-106.doi: 10.11959/j.issn.1000-0801.2021016

基于GAN的无监督域自适应行人重识别

郑声晟, 殷海兵, 黄晓峰, 章天杰

杭州电子科技大学通信工程学院，浙江杭州 310018

修回日期:2020-09-02 出版日期:2021-02-20 发布日期:2021-02-01
作者简介:郑声晟（1996- ），男，杭州电子科技大学通信工程学院硕士生，主要研究方向为行人重识别。
殷海兵（1974- ），男，博士，杭州电子科技大学通信工程学院教授，主要研究方向为数字视频编解码、多媒体信号处理、芯片结构设计验证。
黄晓峰（1988- ），男，博士，杭州电子科技大学通信工程学院讲师，主要研究方向为数字视频编解码与芯片架构设计。
章天杰（2000- ），男，杭州电子科技大学通信工程学院在读，主要研究方向为行人重识别。

GAN-based unsupervised domain adaptive person re-identification

Shengsheng ZHENG, Haibing YIN, Xiaofeng HUANG, Tianjie ZHANG

College of Communication Engineering, Hangzhou Dianzi University, Hangzhou 310018, China

Revised:2020-09-02 Online:2021-02-20 Published:2021-02-01

摘要/Abstract

摘要：

针对无监督域自适应行人重识别中存在的聚类不准确导致网络识别准确率低的问题，提出一种基于生成对抗网络的无监督域自适应行人重识别方法。首先通过在池化层后使用批量归一化层、删除一层全连接层和使用Adam优化器等方法优化CNN模型；然后基于最小错误率贝叶斯决策理论分析聚类错误率和选择聚类关键参数；最后利用生成对抗网络调整聚类，有效提升了无监督域自适应行人重识别的识别准确率。在源域Market-1501和目标域DukeMTMC-reID下进行实验，mAP和Rank-1分别达到了53.7%和71.6%。

关键词: 无监督域自适应, 行人重识别, 生成对抗网络

Abstract:

Aiming at the problem that the inaccurate clustering in the unsupervised domain adaptive pedestrian re-recognition results in low network recognition accuracy, an unsupervised domain adaptive pedestrian re-recognition method based on generative confrontation network was proposed.Firstly, the CNN model was optimized by using the batch normalization layer after the pooling layer, deleting a fully connected layer and adopting the Adam optimizer.Secondly, the cluster error was analyzed and the important parameter in the cluster was decided based on minimum error rate Bayesian decision theory.Finally, the generative adversarial network was utilized to adjust the cluster.These steps effectively improved the recognition accuracy of unsupervised domain adaptive person re-identification.In the case of the source domain Market-1501 and the target domain DukeMTMC-reID, experimental results show that mAP and Rank-1 can reach 53.7% and 71.6%, respectively.

Key words: unsupervised domain adaptive, person re-identification, generative adversarial network

中图分类号:

TP391

郑声晟, 殷海兵, 黄晓峰, 章天杰. 基于GAN的无监督域自适应行人重识别[J]. 电信科学, 2021, 37(2): 99-106.

Shengsheng ZHENG, Haibing YIN, Xiaofeng HUANG, Tianjie ZHANG. GAN-based unsupervised domain adaptive person re-identification[J]. Telecommunications Science, 2021, 37(2): 99-106.

图/表 12

图1

图2

图3

图4

图5

图6

图7

表1

图8

图9

表2

表3

参考文献 18

[1]	ZHENG L , YANG Y , HAUPTMANN A G ,et al. Person re-identification:past,present and future[J]. arXiv:1610.02984, 2012.
[2]	王志宏, 杨震 . 人工智能技术研究及未来智能化信息服务体系的思考[J]. 电信科学, 2017,33(5): 1-11.
	WANG Z H , YANG Z . Research on artificial intelligence technology and the future intelligent information service architecture[J]. Telecommunications Science, 2017,33(5): 1-11.
[3]	杨锋, 许玉, 尹梦晓 ,等. 基于深度学习的行人重识别综述[J]. 计算机应用, 2020,40(5): 1243-1252.
	YANG F , XU Y , YIN M X ,et al. Review on deep learning-based pedestrian re-identification[J]. Journal of Computer Applications, 2020,40(5): 1243-1252.
[4]	罗浩 . 深度学习时代的行人重识别技术[J]. 人工智能, 2019(2): 40-49.
	LUO H . Person re-identification technology in the era of deep learning[J]. Artificial Intelligence, 2019(2): 40-49.
[5]	GOODFELLOW I , POUGETABADIE J , MIRZA M ,et al. Generative adversarial nets[C]// Proceedings of the 27th International Conference on Neural Information Processing Systems-Volume 2. New York:ACM Press, 2014: 2672-2680.
[6]	贾川民, 赵政辉, 王苫社 . 基于神经网络的图像视频编码[J]. 电信科学, 2019,35(5): 32-42.
	JIA C M , ZHAO Z H , WANG S S . Neural network based image and video coding technologies[J]. Telecommunications Science, 2019,35(5): 32-42.
[7]	王万良, 李卓蓉 . 生成式对抗网络研究进展[J]. 通信学报, 2018,39(2): 135-148.
	WANG W L , LI Z R . Advances in generative adversarial network[J]. Journal on Communications, 2018,39(2): 135-148.
[8]	陈亮, 吴攀, 刘韵婷 ,等. 生成对抗网络GAN的发展与最新应用[J]. 电子测量与仪器学报, 2020,34(6): 70-78.
	CHEN L , WU P , LIU Y T ,et al. Development and application of the latest generation against the network of GAN[J]. Journal of Electronic Measurement and Instrumentation, 2020,34(6): 70-78.
[9]	PAN S J , TSANG I W , KWOK J T ,et al. Domain adaptation via transfer component analysis[J]. IEEE Transactions on Neural Networks, 2011,22(2): 199-210.
[10]	CHEN B , LAM W , TSANG I W ,et al. Extracting discriminative concepts for domain adaptation in text mining[C]// Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York:ACM Press, 2009: 179-188.
[11]	NIGAM K , MCCALLUM A , MITCHELL T . Semi-supervised text classification using EM[M]// Semi-supervised learning. Cambridge: MIT Press, 2006: 33-38.
[12]	DENG W , ZHENG L , YE Q ,et al. Image-image domain adaptation with preserved self-similarity and domain-dissimilarity for person re-identification[J]. arXiv:1711.07027, 2018.
[13]	WEI L , ZHANG S , GAO W ,et al. Person transfer GAN to bridge domain gap for person re-identification[C]// Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2018: 79-88.
[14]	WANG J , ZHU X , GONG S ,et al. Transferable joint attribute-identity deep learning for unsupervised person re-identification[C]// Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2018: 2275-2284.
[15]	LI Y , YANG F , LIU Y ,et al. Adaptation and re-identification network:an unsupervised deep transfer learning approach to person re-identification[C]// Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2018: 172-178.
[16]	SONG L , WANG C , ZHANG L ,et al. Unsupervised domain adaptive re-identification:theory and practice[J]. arXiv:1807.11334, 2018.
[17]	MA L , JIA X , SUN Q ,et al. Pose guided person image generation[C]// Proceedings of the 31st International Conference on Neural Information Processing Systems. New York:ACM Press, 2017: 406-416.
[18]	XIONG F , XIAO Y , CAO Z ,et al. Towards good practices on building effective CNN baseline model for person re-identification[J]. arXiv:1807.11042, 2018.

数据集	训练集		测试集
数据集	ID	图像	ID	图像
Market-1501	751	12 936	750	19 732
DukeMTMC-reID	702	16 522	702	19 889

P	mAP	Rank-1	Rank-5	Rank-10
1.0×10^-3	50.3%	69.6%	80.1%	83.8%
1.1×10^-3	50.5%	69.7%	80.5%	84.1%
1.2×10^-3	51.0%	69.8%	81.1%	84.6%
1.3×10^-3	51.5%	69.8%	81.2%	84.9%
1.4×10^-3	51.9%	70.0%	81.2%	84.8%
1.5×10^-3	52.2%	70.2%	81.8%	85.5%
1.6×10^-3	52.0%	70.1%	81.6%	85.2%
1.7×10^-3	51.6%	69.9%	81.3%	85.3%
1.8×10^-3	50.7%	69.7%	80.9%	84.3%
1.9×10^-3	50.2%	69.6%	79.8%	83.7%
2.0×10^-3	49.8%	68.6%	80.3%	83.6%

方法	mAP	Rank-1	Rank-5	Rank-10
Direct Transfer	11.9%	27.3%	41.2%	47.1%
Self-training Baseline	24.7%	40.8%	53.9%	60.5%
SPGAN^[12]	26.2%	46.4%	62.3%	68.0%
TJ-AIDL^[14]	23.0%	44.3%	59.6%	65.0%
ARN^[15]	33.4%	60.2%	73.9%	79.5%
UDA^[16]	49.0%	68.4%	80.1%	83.5%
本文算法	53.7%	71.6%	83.3%	88.2%

基于GAN的无监督域自适应行人重识别

GAN-based unsupervised domain adaptive person re-identification

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 18

相关文章 4

Metrics

推荐阅读 0

[1]	马辉, 王瑞琴, 杨帅. 一种渐进式增长条件生成对抗网络模型[J]. 电信科学, 2023, 39(6): 105-113.
[2]	阿克弘, 胡晓东. 基于GAN数据重构的电信用户流失预测方法[J]. 电信科学, 2023, 39(3): 135-142.
[3]	孙金杨, 刘柏嵩, 任豪, 钱江波. TAGAN：一种融合细粒度语义特征的学术论文对抗推荐算法[J]. 电信科学, 2021, 37(8): 57-65.
[4]	华郁秀, 李荣鹏, 赵志峰, 吴建军, 张宏纲. 基于生成对抗网络的MIMO信道估计方法[J]. 电信科学, 2021, 37(6): 14-22.