Person Re-Identification Based on Squeeze and Excitation Residual Neural Network and Feature Fusion

Ke Wu; Baohua Zhang; Xiaoqi Lü; Yu Gu; Yueming Wang; Xin Liu; Yan Ren; Jianjun Li; Ming Zhang

doi:10.3788/LOP57.181007

Laser & Optoelectronics Progress, Volume. 57, Issue 18, 181007(2020)

Person Re-Identification Based on Squeeze and Excitation Residual Neural Network and Feature Fusion

Ke Wu¹, Baohua Zhang^1、*, Xiaoqi Lü², Yu Gu¹, Yueming Wang¹, Xin Liu¹, Yan Ren¹, Jianjun Li¹, and Ming Zhang¹

¹School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou, Inner Mongolia 0 14010, China

²College of Information Engineering, Inner Mongolia University of Technology, Huhehot, Inner Mongolia 0 10080, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(9)

Fig. 1. Modules of SE-ResNet and ResNet

Download full size

Fig. 2. Activation functions of ReLU and Leaky ReLU. (a) ReLU; (b) Leaky ReLU

Download full size

Fig. 3. Structure diagram of SE-ResNet

Download full size

Fig. 4. Effect diagram of improved SE-ResNet pedestrian re-identification algorithm

Download full size

Fig. 5. Partial pedestrian images in two datasets. (a) DukeMTMC-reID dataset; (b) MarKet-1501 dataset

Download full size

Fig. 6. Rank-k and mAPs at different fusion schemes. (a) Rank-k; (b) mAP

Download full size

Table 1. Experimental results of different convolution sizes

View table

Table 1. Experimental results of different convolution sizes

Convolution kernel size	Rank-1 /%	Rank-5 /%	Rank-10 /%	mAP /%	Running time /s
3×3	91.60	95.60	96.40	87.80	1316
5×5	92.10	95.40	96.30	87.90	1807
7×7	93.10	96.00	97.00	89.00	1983
9×9	91.74	95.72	97.00	87.59	12273
11×11	91.74	95.72	97.00	87.59	14146

Table 2. Performance comparison of different algorithms (Market-1501 dataset)unit: %

View table

Table 2. Performance comparison of different algorithms (Market-1501 dataset)unit: %

Algorithm	Rank-1	Rank-5	Rank-10	mAP
DenseNet-121^[13]	90.17	94.50	96.08	74.02
PCB^[14]	92.64	93.37	94.95	77.47
ResNet50	88.84	94.84	96.62	71.59
SE-ResNet(before fusion)	88.53	93.28	94.89	82.25
SE-ResNet(after fusion)	93.10	96.00	97.00	89.00

Table 3. Performance comparison of different algorithms (DukeMTMC-reID dataset) unit: %

View table

Table 3. Performance comparison of different algorithms (DukeMTMC-reID dataset) unit: %

Algorithm	Rank-1	Rank-5	Rank-10	mAP
PCB	85.68	91.47	93.49	80.68
DenseNet-121	80.02	88.86	91.74	63.63
ResNet50	85.59	91.38	92.26	80.65
SE-ResNet(before fusion)	57.94	69.70	74.59	46.75
SE-ResNet(after fusion)	86.00	91.42	94.03	81.24

Tools

Get Citation

Copy Citation Text

Ke Wu, Baohua Zhang, Xiaoqi Lü, Yu Gu, Yueming Wang, Xin Liu, Yan Ren, Jianjun Li, Ming Zhang. Person Re-Identification Based on Squeeze and Excitation Residual Neural Network and Feature Fusion[J]. Laser & Optoelectronics Progress, 2020, 57(18): 181007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Dec. 24, 2019

Accepted: Feb. 10, 2020

Published Online: Sep. 2, 2020

The Author Email: Baohua Zhang (zbh_wj2004@imust.cn)

DOI:10.3788/LOP57.181007

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology