基于多标签对抗领域自适应的行人属性识别算法

胡强梁; 陈琳; 尚明生

doi:10.13700/j.bh.1001-5965.2023.0386

基于多标签对抗领域自适应的行人属性识别算法

doi: 10.13700/j.bh.1001-5965.2023.0386

胡强梁^{1, 2},
陈琳^{1, 2, ,},
尚明生^{1, 2}

1.
中国科学院重庆绿色智能技术研究院电子信息所，重庆 400714
2.
中国科学院大学重庆学院人工智能学院，重庆 400714

基金项目:

国家自然科学基金（82030066,61902370）；重庆市教委在渝高校与中科院所合作重点项目（HZ2021008,HZ2021017）

详细信息

通讯作者:
E-mail：chenlin@cigit.ac.cn

中图分类号: V351；TB391
计量
- 文章访问数: 393
- HTML全文浏览量: 96
- PDF下载量: 14
- 被引次数: 0
出版历程
- 收稿日期: 2023-06-16
- 录用日期: 2023-09-23
- 网络出版日期: 2023-11-29
- 整期出版日期: 2025-07-31

Pedestrian attribute recognition algorithm based on multi-label adversarial domain adaptation

HU Qiangliang^{1, 2},
CHEN Lin^{1, 2
, ,},
SHANG Mingsheng^{1, 2}

1.
Chongqing Institute of Green and Intelligent Technology，Chinese Academy of Sciences，Chongqing 400714，China
2.
School of Artificial Intelligence，Chongqing School of University of Chinese Academy of Sciences，Chongqing 400714，China

Funds:

National Natural Science Foundation of China (82030066,61902370); The Key Project of Chongqing Municipal Education Commission based on Cooperation of Chongqing Local Universities and the Chinese Academy of Sciences (HZ2021008,HZ2021017)

More Information

Corresponding author: E-mail：chenlin@cigit.ac.cn

摘要

摘要:
针对无监督领域自适应算法通常局限于单标签学习问题，难以适配针对行人属性的多标签分类任务，提出一种多标签对抗领域自适应的行人属性识别算法。为适应行人属性多标签领域迁移任务，基于多标签特征分离模块，利用特定类别语义对主干网络提取的深度特征进行属性分离，有效提取特定属性的表征信息。针对不同领域属性特征分布差异较大的难点，提出基于分类器复用的多标签领域鉴别模块，同时实现多标签领域对齐和多标签分类，有效利用预测的鉴别信息捕获特征分布的多模式结构。实验结果表明：所提算法对比基准模型有明显提升，在平均准确率、准确率、召回率和F1指标上分别提升了4.49%、5.5%、11.44%和5.89%；所提算法为多标签领域自适应学习提供了新思路。
- 行人属性识别 /
- 多标签学习 /
- 多标签领域自适应 /
- 多属性特征提取 /
- 多标签分类
Abstract:
Current unsupervised domain adaption algorithms usually consider only single-label learning, failing to adapt to multi-label classification tasks in pedestrian attribute recognition. To address this issue, a multi-label adversarial domain adaptation algorithm was proposed for pedestrian attribute recognition. First, to adapt to the domain transfer task of multiple attribute labels, a multi-label feature disentanglement module was employed to effectively disentangle attributes of deep features extracted from the backbone network with specific attributes based on category-specific semantics. Secondly, to reduce the gap of attribute feature distributions in different domains, a multi-label domain discrimination module based on classifier reuse was proposed to achieve both multi-attribute domain alignment and multi-label classification, which effectively exploited the predicted discrimination information to capture the multi-mode structures of feature distribution. Experimental results show that the proposed algorithm achieves superior results than the baseline model, with improvements of 4.49%, 5.5%, 11.44% and 5.89% on mean accuracy, accuracy, recall and F1 values, respectively. The proposed algorithm provides a new insight for multi-label domain adaptation learning.
- pedestrian attribute recognition /
- multi-label learning /
- multi-label domain adaptation /
- multi-label feature extraction /
- multi-label classification

HTML全文

图 1 本文算法框架

Figure 1. Framework of the proposed algorithm

下载: 全尺寸图片幻灯片

图 2 对比实验效果

Figure 2. Comparison of experiment results

下载: 全尺寸图片幻灯片

图 3 RAPv1$ \Rrightarrow $PA100k的各属性准确率

Figure 3. Accuracy of each attribute on RAPv1$ \Rrightarrow $PA100k

下载: 全尺寸图片幻灯片

图 4 RAPv2$ \Rrightarrow $PA100k的各属性准确率

Figure 4. Accuracy of each attribute on RAPv2$ \Rrightarrow $PA100k

下载: 全尺寸图片幻灯片

图 5 属性识别效果分析（RAPv2$ \Rrightarrow $PA100k）

Figure 5. Performance of attribute recognition on RAPv2$ \Rrightarrow $PA100k

下载: 全尺寸图片幻灯片

表 1 实验软硬件环境

Table 1. Experimental hardware and software environment

实验环境	配置
CPU	Intel Xeon E5-2680 v4
GPU	NVIDIA GeForce RTX 1080Ti
操作系统	Ubuntu 18.04.2
深度学习框架	Pytorch 1.9
计算机语言	Python 3.8

下载: 导出CSV

表 2 模型参数设置

Table 2. Model parameters setting

参数	数值
图片预处理大小	256×192
批处理大小	64
学习率	0.01
动量大小	0.9
训练轮次	30

下载: 导出CSV

表 3 在RAPv1$ \Rrightarrow $PA100k上的消融实验

Table 3. Ablation experiments on RAPv1$ \Rrightarrow $PA100k

算法	E_mA/%	E_Acc/%	E_Prec/%	E_Rec/%	E_F1/%
ResNet50	64.85	42.22	65.75	49.71	53.44
ResNet50+MFD	65.32	43.04	67.17	51.93	54.99
ResNet50+MFD+DA	65.83	39.41	63.43	45.86	50.01
ResNet50+MDD	66.34	42.14	66.95	47.99	52.99
本文算法	69.34	47.72	64.33	61.15	59.33

下载: 导出CSV

表 4 在RAPv2$ \Rrightarrow $PA100k上的消融实验

Table 4. Ablation experiments on RAPv2$ \Rrightarrow $PA100k

算法	E_mA/%	E_Acc/%	E_Prec/%	E_Rec/%	E_F1/%
ResNet50	67.19	47.24	68.93	56.01	58.47
ResNet50+MFD	68.82	47.92	73.15	55.28	59.57
ResNet50+MFD+DA	67.74	44.51	66.22	53.26	55.66
ResNet50+MDD	64.34	41.14	67.56	46.15	51.88
本文算法	72.04	47.99	62.85	64.01	60.33

下载: 导出CSV

参考文献(33)

[1]	WANG X, ZHENG S F, YANG R, et al. Pedestrian attribute recognition: a survey[J]. Pattern Recognition, 2022, 121: 108220. doi: 10.1016/j.patcog.2021.108220
[2]	WILSON G, COOK D J. A survey of unsupervised deep domain adaptation[J]. ACM Transactions on Intelligent Systems and Technology, 2020, 11(5): 1-46.
[3]	STACKE K, EILERTSEN G, UNGER J, et al. Measuring domain shift for deep learning in histopathology[J]. IEEE Journal of Biomedical and Health Informatics, 2021, 25(2): 325-336. doi: 10.1109/JBHI.2020.3032060
[4]	YU H X, ZHENG W S, WU A, et al. Unsupervised person re-identification by soft multilabel learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2019: 2148-2157.
[5]	WANG D, ZHANG S. Unsupervised person re-identification via multi-label classification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 10978-10987.
[6]	LI D W, CHEN X T, HUANG K Q. Multi-attribute learning for pedestrian attribute recognition in surveillance scenarios[C]//Proceedings of 3rd IAPR Asian Conference on Pattern Recognition. Piscataway: IEEE Press, 2015: 111-115.
[7]	SARFRAZ M S, SCHUMANN A, WANG Y, et al. Deep view-sensitive pedestrian attribute inference in an end-to-end model[C]//Proceedings of the British Machine Vision Conference. London: BMVA Press, 2017, 134: 1-13.
[8]	YU K, LENG B, ZHANG Z, et al. Weakly-supervised learning of mid-level features for pedestrian attribute recognition and localization[C]//Proceedings of the British Machine Vision Conference. London: BMVA Press, 2017: 1-11.
[9]	LIU P, LIU X, YAN J, et al. Localization guided learning for pedestrian attribute recognition[C]//Proceedings of the British Machine Vision Conference. London: BMVA Press, 2018: 142-152.
[10]	LI D, CHEN X, ZHANG Z, et al. Pose guided deep model for pedestrian attribute recognition in surveillance scenarios[C]//Proceedings of the IEEE International Conference on Multimedia and Expo. Piscataway: IEEE Press, 2018: 1-6.
[11]	WU M D, HUANG D, GUO Y F, et al. Distraction-aware feature learning for human attribute recognition via coarse-to-fine attention mechanism[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2020: 12394-12401.
[12]	ZHANG J, REN P, LI J. Deep template matching for pedestrian attribute recognition with the auxiliary supervision of attribute-wise keypoints[EB/OL]. (2020-11-13)[2020-05-26]. http://doi.org/10.48550/arXiv.2011.06798.
[13]	LIU X, ZHAO H, TIAN M, et al. Hydraplus-net: attentive deep features for pedestrian analysis[C]//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE Press, 2017: 350-359.
[14]	GUO H, ZHENG K, FAN X, et al. Visual attention consistency under image transforms for multi-label image classification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2019: 729-739.
[15]	JI Z, HE E, WANG H, et al. Image-attribute reciprocally guided attention network for pedestrian attribute recognition[J]. Pattern Recognition Letters, 2019, 120: 89-95. doi: 10.1016/j.patrec.2019.01.010
[16]	TAN Z, YANG Y, WAN J, et al. Attention-based pedestrian attribute analysis[J]. IEEE Transactions on Image Processing, 2019, 28(12): 6126-6140. doi: 10.1109/TIP.2019.2919199
[17]	ZHAO H, TIAN M, SUN S, et al. Spindle net: person re-identification with human body region guided feature decomposition and fusion[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2017: 1077-1085.
[18]	LI Q, ZHAO X, HE R, et al. Visual-semantic graph reasoning for pedestrian attribute recognition[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2019, 33(1): 8634-8641.
[19]	WU J, LIU H, JIANG J, et al. Person attribute recognition by sequence contextual relation learning[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2020, 30(10): 3398-3412. doi: 10.1109/TCSVT.2020.2982962
[20]	LONG M, CAO Y, WANG J, et al. Learning transferable features with deep adaptation networks[C]//Proceedings of the International Conference on Machine Learning. Piscataway: IEEE Press, 2015: 97-105.
[21]	LONG M, ZHU H, WANG J, et al. Deep transfer learning with joint adaptation networks[C]//Proceedings of the International Conference on Machine Learning. Piscataway: IEEE Press, 2017: 2208-2217.
[22]	GANIN Y, LEMPITSKY V. Unsupervised domain adaptation by backpropagation[C]//Proceedings of the International Conference on Machine Learning. Piscataway: IEEE Press, 2015, 9908: 1180-1189.
[23]	GHIFARY M, KLEIJN W B, ZHANG M, et al. Deep reconstruction-classification networks for unsupervised domain adaptation[C]//Proceedings of the European Conference on Computer Vision. Berlin: Springer, 2016: 597-613.
[24]	BOUSMALIS K, TRIGEORGIS G, SILBERMAN N, et al. Domain separation networks[C]//Proceedings of the Annual Conference on Neural Information Processing Systems. Piscataway: IEEE Press, 2016, 29: 343-351.
[25]	TZENG E, HOFFMAN J, SAENKO K, et al. Adversarial discriminative domain adaptation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2017: 7167-7176.
[26]	PEI Z, CAO Z, LONG M, et al. Multi-adversarial domain adaptation[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2018, 32(1): 3934-3941.
[27]	HOFFMAN J, TZENG E, PARK T, et al. Cycada: cycle-consistent adversarial domain adaptation[C]//Proceedings of the International Conference on Machine Learning. Piscataway: IEEE Press, 2018: 1989-1998.
[28]	LONG M, CAO Z, WANG J, et al. Conditional adversarial domain adaptation[C]//Proceedings of the Annual Conference on Neural Information Processing Systems. Piscataway: IEEE Press, 2018, 31: 1-11.
[29]	SAITO K, WATANABE K, USHIKU Y, et al. Maximum classifier discrepancy for unsupervised domain adaptation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 3723-3732.
[30]	CHEN T, XU M, HUI X, et al. Learning semantic-specific graph representation for multi-label image recognition[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE Press, 2019: 522-531.
[31]	CHEN L, CHEN H, WEI Z, et al. Reusing the task-specific classifier as a discriminator: discriminator-free adversarial domain adaptation[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2022: 7181-7190.
[32]	JIA J, HUANG H, YANG W, et al. Rethinking of pedestrian attribute recognition: realistic datasets with efficient method[EB/OL]. (2020-05-25)[2020-05-26]. http://doi.org/10.48550/arXiv.2005.11909.
[33]	LI D, ZHANG Z, CHEN X, et al. A richly annotated pedestrian dataset for person retrieval in real surveillance scenarios[J]. IEEE Transactions on Image Processing, 2018, 28(4): 1575-1590.