基于多尺度特征的行人重识别属性提取新方法

doi:10.13229/j.cnki.jdxbgxb.20220327

摘要/Abstract

摘要：

针对目前行人重识别存在背景混乱、遮挡干扰、姿势不对齐、解释性不足等问题，提出了一种基于多尺度特征的行人重识别属性提取新方法。通过中间层学习模块学习属性并按属性分组卷积，以增强模型的可解释性；通过选择对比中间的属性特征，减少干扰特征的影响。本文方法在3个常用的公开数据集上测试并与其他方法进行了对比，实验结果表明，该方法可以有效提取中间层的语义信息，并且在跨数据集上测试优于其他方法。

关键词: 计算机应用, 可解释性, 行人重识别, 多尺度特征

Abstract:

At present， person re-identification （ReID） still exists some issues to solve， such as background confusion， blocking interference， pose misalignment， and insufficient explanation. In this paper， a new method based on multi-scale feature extraction is proposed for person ReID. Specifically， using the feature learning module to learn attributes and convolving by attribute grouping， which could enhance the interpretability of the model and reduce the impact of interference features by comparing the intermediate property characteristics. Compared with other methods， the new method was tested on three commonly used public datasets. The detailed experiments demonstrate that the new method can efficiently extract semantic information from the middle layer and outperforms other methods tested across datasets.

Key words: computer application, interpretability, person re-identification, multi-scale features

中图分类号:

TP391.4

于鹏,朴燕. 基于多尺度特征的行人重识别属性提取新方法[J]. 吉林大学学报(工学版), 2023, 53(4): 1155-1162.

Peng YU,Yan PIAO. New method for extracting person re-identification attributes based on multi-scale features[J]. Journal of Jilin University(Engineering and Technology Edition), 2023, 53(4): 1155-1162.

图/表 8

图1

表1

图2

表2

图3

表3

表4

表5

参考文献 27

1	Alex K, Ilya S, Geoffrey E H. Imagenet classification with deep convolutional neural networks[J]. Advances in neural information processing systems, 2017, 60(6): 84-90.
2	Zhang Zhi-zheng, Lan Cui-ling, Zeng Wen-jun, et al. Relation-aware global attention for person re-identification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, USA, 2020: 3186-3195.
3	Chen Guang-yi, Lin Chun-ze, Ren Liang-liang, et al. Self-critical attention learning for person re-identification[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, Korea (South), 2019: 9637-9646.
4	Fang Peng-fei, Zhou Jie-ming, Kumar Roy Soumava, et al. Bilinear attention networks for person retrieval[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, Korea (South), 2019: 8030-8039.
5	Sun Yi-fan, Zheng Liang, Yang Yi, et al. Beyond part models: person retrieval with refined part pooling (and a strong convolutional baseline)[C]//Proceedings of the European Conference on Computer Vision (ECCV), Munich, Germany, 2018: 501-518.
6	Zhou Kai-yang, Yang Yong-xin, Cavallaro Andrea, et al. Omni-scale feature learning for person re-identification[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, Korea (South), 2019: 3702-3712.
7	Chen Xiao-dong, Liu Xin-chen, Liu Wu, et al. Explainable person re-identification with attribute-guided metric distillation[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Montreal, Canada, 2021: 11813-11822.
8	Divyansh G, Netra P, Thomas T, et al. Towards explainable person re-identification[C]//2021 IEEE Symposium Series on Computational Intelligence (SSCI), Orlando, USA, 2021: 1-8.
9	Chen Guang-yi, Gu Tian-pei, Lu Ji-wen, et al. Person re-identification via attention pyramid[J]. IEEE Transactions on Image Processing, 2021, 30: 7663-7676.
10	Wang Guan-shuo, Yuan Yu-feng, Chen Xiong, et al. Learning discriminative features with multiple granularities for person re-identification[C]//Proceedings of the 26th ACM International Conference on Multimedia, Seoul, Korea(South), 2018: 274-282.
11	Feng Zheng, Cheng Deng, Xing Sun, et al. Pyramidal person re-identification via multi-loss dynamic training[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 8514-8522.
12	本杰明•里贝特. 心智时间: 意识中的时间因素[M]: 李恒熙,李恒威,罗慧怡译.杭州: 浙江大学出版社, 2013:55-56.
13	Han Yi-zeng, Huang Gao, Song Shi-ji, et al. Dynamic neural networks: a survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(11): 7436-7456.
14	Teerapittayanon S, McDanel B, Kung H T. Branchynet: fast inference via early exiting from deep neural networks[C]//The 23rd International Conference on Pattern Recognition (ICPR), Cancun, Mexico, 2016: 2464-2469.
15	Hermans Alexander, Beyer Lucas, Leibe Bastian. In defense of the triplet loss for person re-identification. arXiv 2017[J/OL].[2017-04-01]. .
16	Zheng Liang, Shen Li-yue, Tian Lu, et al. Scalable person re-identification: a benchmark[C]//Proceedings of the IEEE International Conference on Computer Vision, Santiago, Chile, 2015: 1116-1124.
17	Zhong Zhun, Zheng Liang, Cao Dong-lin, et al. Re-ranking person re-identification with k-reciprocal encoding[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 1318-1327.
18	Ergys R, Francesco S, Roger Z, et al. Performance measures and a data set for multi-target, multi-camera tracking[C]//European Conference on Computer Vision, Amsterdam, The Netherlands, 2016: 17-35.
19	Lin Yu-tian, Zheng Liang, Zheng Zhe-dong, et al. Improving person re-identification by attribute and identity learning[J]. Pattern Recognition, 2019, 95: 151-161.
20	Jia Deng, Wei Dong, Socher Richard, et al. Imagenet: a large-scale hierarchical image database[C]//IEEE Conference on Computer Vision and Pattern Recognition, Miami, USA, 2009: 248-255.
21	Hyunjong P, Bumsub H. Relation network for person re-identification[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 11839-11847.
22	Sun Yi-fan, Xu Qin, Li Ya-li, et al. Perceive where to focus: learning visibility-aware part-level features for partial person re-identification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 393-402.
23	Tay C-P, Roy S, Yap K-H. Aanet: attribute attention network for person re-identifications[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 7134-7143.
24	Yang Wen-jie, Huang Hou-jing, Zhang Zhang, et al. Towards rich feature discovery with class activation maps augmentation for person re-identification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 1389-1398.
25	Chen Bing-hui, Deng Wei-hong, Hu Jia-ni. Mixed high-order attention network for person re-identification[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, Korea (South), 2019: 371-381.
26	Chen Xue-song, Fu Can-miao, Zhao Yong, et al. Salience-guided cascaded suppression network for person re-identification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, USA, 2020: 3300-3310.
27	Zhao Shi-zhen, Gao Chang-xin, Zhang Jun, et al. Do not disturb me: person re-identification under the interference of other pedestrians[C]//European Conference on Computer Vision, Glasgow, UK, 2020: 647-663.

相关文章 15

[1]	姜宇,潘家铮,陈何淮,符凌智,齐红. 基于分割方法的繁体中文报纸文本检测[J]. 吉林大学学报(工学版), 2023, 53(4): 1146-1154.
[2]	吴振宇,刘小飞,王义普. 基于DKRRT^*-APF算法的无人系统轨迹规划[J]. 吉林大学学报(工学版), 2023, 53(3): 781-791.
[3]	陶博,颜伏伍,尹智帅,武冬梅. 基于高精度地图增强的三维目标检测算法[J]. 吉林大学学报(工学版), 2023, 53(3): 802-809.
[4]	薛珊,张亚亮,吕琼莹,曹国华. 复杂背景下的反无人机系统目标检测算法[J]. 吉林大学学报(工学版), 2023, 53(3): 891-901.
[5]	潘弘洋,刘昭,杨波,孙庚,刘衍珩. 基于新一代通信技术的无人机系统群体智能方法综述[J]. 吉林大学学报(工学版), 2023, 53(3): 629-642.
[6]	何颖,樊俊松,王巍,孙庚,刘衍珩. 无人机空地安全通信与航迹规划的多目标联合优化方法[J]. 吉林大学学报(工学版), 2023, 53(3): 913-922.
[7]	郭鹏,赵文超,雷坤. 基于改进Jaya算法的双资源约束柔性作业车间调度[J]. 吉林大学学报(工学版), 2023, 53(2): 480-487.
[8]	刘近贞,高国辉,熊慧. 用于脑组织分割的多尺度注意网络[J]. 吉林大学学报(工学版), 2023, 53(2): 576-583.
[9]	祁贤雨,王巍,王琳,赵玉飞,董彦鹏. 基于物体语义栅格地图的语义拓扑地图构建方法[J]. 吉林大学学报(工学版), 2023, 53(2): 569-575.
[10]	时小虎,吴佳琦,吴春国,程石,翁小辉,常志勇. 基于残差网络的弯道增强车道线检测方法[J]. 吉林大学学报(工学版), 2023, 53(2): 584-592.
[11]	曲福恒,丁天雨,陆洋,杨勇,胡雅婷. 基于邻域相似性的图像码字快速搜索算法[J]. 吉林大学学报(工学版), 2022, 52(8): 1865-1871.
[12]	白天,徐明蔚,刘思铭,张佶安,王喆. 基于深度神经网络的诉辩文本争议焦点识别[J]. 吉林大学学报(工学版), 2022, 52(8): 1872-1880.
[13]	周丰丰,朱海洋. 基于三段式特征选择策略的脑电情感识别算法SEE[J]. 吉林大学学报(工学版), 2022, 52(8): 1834-1841.
[14]	赵宏伟,张健荣,朱隽平,李海. 基于对比自监督学习的图像分类框架[J]. 吉林大学学报(工学版), 2022, 52(8): 1850-1856.
[15]	秦贵和,黄俊锋,孙铭会. 基于双手键盘的虚拟现实文本输入[J]. 吉林大学学报(工学版), 2022, 52(8): 1881-1888.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

阶段		输出尺寸	OSNet	阶段	输出尺寸	MLLANet
	conv1	128×64， 64	7×7 conv， stride 2
	conv1	64×32， 64	3×3 max pool， stride 2
1层	conv2	64×32， 256	bottleneck × 2
	transition	64×32， 256	1×1 conv
	transition	32×16， 256	2×2 average pool， stride 2
	MLL0	32×16， 256	1×1 conv	MLLA0	32×16， 256×30+30	1×1 conv
		1×1， 256	global average pool		1×1， 256×30+30	global average pool
		1×1， 256	fc		1×1， 256×30+30	fc
2层	conv3	32×16， 384	bottleneck × 2
	transition	32×16， 384	1×1 conv
	transition	16×8， 384	2×2 average pool， stride 2
	MLL1	16×8， 384	1×1 conv	MLLA1	16×8， 384 ×30+30	1×1 conv
		1×1， 384	global average pool		1×1， 384 ×30+30	global average pool
		1×1， 384	fc		1×1， 384 ×30+30	fc
3层	conv4	16×8， 512	bottleneck × 2
	MLL2	16×8， 512	1×1 conv	MLLA2	16×8， 512 ×30+30	1×1 conv
		1×1， 512	global average pool		1×1， 512 ×30+30	global average pool
		1×1， 512	fc		1×1， 512 ×30+30	fc

	2层	3层	4层
飞机	313	415	272
汽车	340	403	257
鸟	338	418	244
猫	303	457	240
鹿	329	436	235
狗	307	460	233
蛙	325	419	256
马	338	415	247
船	295	450	255
卡车	312	427	261

模型	权重	平均精度	前1次命中的准确率	前5次命中的准确率
OSNet	预训练	84.9	94.8	96.8
OSNet	无预训练	63.0	83.5	93.3
OSNet+MLL	预训练	86.7	95.9	97.4
OSNet+MLL	无预训练	72.5	89.3	95.5
GCP	预训练	88.9	95.2	98.2
GCP	无预训练	85.3	93.9	97.9
GCP+MLL	预训练	89.7	95.4	98.5
GCP+MLL	无预训练	87.7	94.4	98.1

方法	主干网	Market?1501		DukeMTMC?ReID			CUHK03
方法	主干网	mAP	Rank?1	mAP	Rank?1	mAP	Rank?1
PCB+RPP^［5］	ResNet?50	81.6	93.8	69.2	83.3	57.5	63.7
MGN^［10］	ResNet?50	86.9	95.6	78.4	88.7	66.0	66.8
VMP^［22］	ResNet?50	80.8	93.0	72.6	83.6	-	-
OSNet^［6］	OSNet	84.9	94.8	73.5	88.6	67.8	72.3
AAnet^［23］	ResNet?50	82.5	93.9	72.6	86.4	-	-
CAMA^［24］	ResNet?50	84.5	94.7	-	-	64.2	66.6
BAT?net^［4］	ResNet?50	85.5	94.1	77.3	87.7	73.2	76.2
MHN^［25］	ResNet?50	85.0	95.1	77.2	89.1	65.4	71.7
SCSN^［26］	ResNet?50	88.5	95.7	79.0	90.1	80.2	84.1
GCP^［21］	ResNet?50	88.9	95.2	78.6	89.7	75.6	77.9
RGA?SC^［2］	ResNet?50	88.4	96.1	-	-	74.5	79.6
?RGA?S^［2］	ResNet?50	88.0	95.7	77.5	88.8	72.7	78.1
PISNet^［27］	ResNet?50	87.1	95.6	78.7	88.8	-	-
APNet?S^［9］	ResNet?50	89.0	96.1	78.8	89.3	78.1	80.9
APNet?C^［9］	ResNet?50	90.5	96.2	81.5	90.4	81.5	83.0
MLLNet	OSNet	86.7	95.9	74.6	89.2	69.7	73.4
MLLANet	OSNet	87.8	96.1	75.1	89.5	-	-
MLLNet2	APNet?C	89.9	95.9	81.7	90.5	81.7	83.1
MLLNet3	GCP	89.7	95.4	79.5	90.0	76.9	78.7

训练集	测试集	模型	平均精度	Rank?1	Rank?5	Rank?10
Market	Duke	APNet?c3	22.8	37.7	52.4	59.0
Market	Duke	OSNet	26.7	48.5	62.3	67.4
Market	Duke	OSNet_MLL	26.9	49.5	63.1	68.0
Market	Duke	OSNet_MLLA	27.1	49.6	63.7	69.0
Duke	Market	APNet?c3	23.7	50.9	66.6	72.6
Duke	Market	OSNet	26.1	57.7	73.7	80.0
Duke	Market	OSNet_MLL	26.2	58.0	73.4	82.5
Duke	Market	OSNet_MLLA	26.8	59.8	75.6	81.5