Researching | Person Re-Identification Based on Feature Stitching

Laser & Optoelectronics Progress, Volume. 56, Issue 16, 162001(2019)

Person Re-Identification Based on Feature Stitching

Tong Pan and Wenguo Li^*

Author Affiliations

Faculty of Mechanical & Electrical Engineering, Kunming University of Science and Technology, Kunming, Yunnan 650500, China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (11)

References (23)

Paper Information

Figures & Tables(11)

Fig. 1. Inception block

Download full size

Fig. 2. Network structure

Download full size

Fig. 3. Recognition effects

Download full size

Table 1. Experimental results corresponding todifferent N%
View table
Table 1. Experimental results corresponding todifferent N%
N Rank1 mAP
2048 94.2 84.9
1024 94.8 85.7
512 95.2 86.0
256 94.6 85.5

Table 2. Experimental results corresponding to different M%
View table
Table 2. Experimental results corresponding to different M%
M Rank1 mAP
1.0 95.2 86.0
0.8 94.5 85.6
0.6 94.6 85.8
0.4 94.3 85.3
0.2 94.0 85.4

Table 3. Experiment results of basic model%
View table
Table 3. Experiment results of basic model%
Method Rank1 mAP
1 (Resnet50) 87.2 72.3
2 (Resnet50+bottleneck layer) 89.6 73.4
3 (Resnet50+bottlenecklayer+bias removing) 90.5 75.1

Table 4. Experimental results of different methods%
View table
Table 4. Experimental results of different methods%
Method Rank1 mAP
E(branch 1) 90.5 75.1
F(branches 1,2) 93.8 83.3
H(branches 1,2,3) 93.9 84.2
J(branches 1,2,3 andauxiliary classifier) 95.2 86.0

Table 5. Test time on Market1501 dataset
View table
Table 5. Test time on Market1501 dataset
Method Test time /s
Resnet50 179
Ours 206
PCB+RPP^[11] 212

Table 6. Comparison of results on Market1501%
View table
Table 6. Comparison of results on Market1501%
Method Rank1 Rank5 Rank10 mAP
MLFN^[16] 90.0 - - 74.3
PSE+ECN^[17] 90.4 94.5 96.8 80.5
HA-CNN^[18] 91.2 - - 75.7
MACS^[7] 93.1 - - 82.3
DNN_CRF^[19] 93.5 - - 81.6
HSP^[12] 93.7 97.6 98.4 83.4
PCB+RPP^[11] 93.8 97.5 98.5 81.6
Ours 95.2 98.0 98.8 86.0

Table 7. Comparison of results on DukeMTMC-Reid%
View table
Table 7. Comparison of results on DukeMTMC-Reid%
Method Rank1 Rank5 Rank10 mAP
AOS^[20] 79.2 - - 62.1
HA-CNN^[18] 80.5 - - 63.8
GSRW^[21] 80.7 88.5 90.8 66.4
MLFN^[16] 81.0 - - 62.8
DuATM^[22] 81.8 90.2 - 64.6
PCB+RPP^[11] 83.3 90.5 92.5 69.2
Ours 86.7 93.5 95.5 74.0

Table 8. Comparison of results on CUHK03%
View table
Table 8. Comparison of results on CUHK03%
Method Rank1 Rank5 Rank10 mAP
HA-CNN^[18] 41.7 - - 38.6
MLFN^[16] 52.8 - - 47.8
PCB+RPP^[11] 63.7 80.6 86.9 57.5
Ours 68.3 84.7 90.5 64.6

Tools

Get Citation

Copy Citation Text

Tong Pan, Wenguo Li. Person Re-Identification Based on Feature Stitching[J]. Laser & Optoelectronics Progress, 2019, 56(16): 162001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Optics in Computing

Received: Feb. 19, 2019

Accepted: Mar. 21, 2019

Published Online: Aug. 5, 2019

The Author Email: Wenguo Li (475284191@qq.com)

DOI:10.3788/LOP56.162001

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology