Multi-Granularity and Cross-Modality Pedestrian Re-Identification Algorithm

Wei Xiong; Ling Yue; Lei Zhou; Kai Zhang; Lirong Li

doi:10.3788/LOP202259.2220001

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 22 >Page 2220001 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 22, 2220001 (2022)

Multi-Granularity and Cross-Modality Pedestrian Re-Identification Algorithm

Wei Xiong^1,2,*, Ling Yue¹, Lei Zhou¹, Kai Zhang¹, and Lirong Li¹

Author Affiliations

¹School of Electrical & Electronic Engineering, Hubei University of Technology, Wuhan 430068, Hubei , China

²Dept. of Computer Science & Engineering, University of South Carolina, Columbia 29201, SC, USA

show less

DOI: 10.3788/LOP202259.2220001 Cite this Article Set citation alerts

Wei Xiong, Ling Yue, Lei Zhou, Kai Zhang, Lirong Li. Multi-Granularity and Cross-Modality Pedestrian Re-Identification Algorithm[J]. Laser & Optoelectronics Progress, 2022, 59(22): 2220001 Copy Citation Text

show less

Fig. 1. Block diagram of proposed algorithm

Download full size

Fig. 2. Non_local attention module

Download full size

Fig. 3. SYSU-MM01 dataset

Download full size

Method	All Search		Indoor Search
Method	Rank-1	mAP	Rank-1	mAP
Baseline	57.34	55.13	63.11	69.20
Baseline+Non_local	61.16	57.03	67.48	72.22
Baseline+Non_local+Branch 1	62.83	58.10	68.89	73.47

Table 1. Ablation experiment

Method	All Search			Indoor Search
Method	Rank-1	mAP	mINP	Rank-1	mAP	mINP
One-stream^［18］（ICCV2017）	12.04	13.67		16.94	22.95
Two-stream^［18］（ICCV2017）	11.65	12.85		15.60	21.49
Zero-padding^［18］（ICCV2017）	14.80	15.95		20.58	26.92
HCML^［6］（AAAI2018）	14.32	16.16		24.52	30.08
cmGAN^［8］（IJCAI2018）	26.97	31.49		31.63	42.19
BDTR^［19］（TIFS2019）	27.32	27.32		31.92	41.86
D²RL^［20］（CVPR2019）	28.90	29.20
MAC^［21］（TIP2020）	33.26	36.22		36.43	37.03
AlignGAN^［22］（ICCV2019）	42.40	40.70		45.90	54.30
Xmodal^［10］（AAAI2020）	49.92	50.73
DDAG^［23］（ECCV2020）	54.75	53.02		61.20	67.98
AGW^［13］（TPAMI2021）	47.50	47.65	35.30	54.17	62.97	59.23
DF²AM^［14］（Arxiv2021）	56.93	55.10		66.39	71.52
DGTL^［17］（ISPL2021）	57.34	55.13		63.11	69.20
Proposed method	62.83	58.10	45.62	68.89	73.47	68.06

Table 2. Comparison results on SYSU-MM01 dataset

Method	Visible to Infrared			Infrared to Visible
Method	Rank-1	mAP	mINP	Rank-1	mAP	mINP
HCML^［6］（AAAI2018）	24.44	20.08		21.70	22.24
Zero-padding^［18］（ICCV2017）	17.75	18.90		16.63	17.82
BDTR^［19］（TIFS2019）	33.56	32.76		32.92	31.96
D²RL^［20］（CVPR2019）	43.40	44.10
MAC^［21］（TIP2020）	36.43	37.03		36.20	36.63
AlignGAN^［22］（ICCV2019）	57.90	53.60		56.30	53.40
Xmodal^［10］（AAAI2020）	62.21	60.18
DDAG^［23］（ECCV2020）	69.34	63.46		68.06	61.80
AGW^［13］（TPAMI2021）	70.05	66.37	50.19
DF²AM^［14］（Arxiv2021）	73.06	67.81		70.49	63.85
DGTL^［17］（ISPL2021）	83.92	73.78		81.59	71.65
Proposed method	87.78	76.22	58.48	84.97	73.81

Table 3. Comparison results on RegDB dataset

Wei Xiong, Ling Yue, Lei Zhou, Kai Zhang, Lirong Li. Multi-Granularity and Cross-Modality Pedestrian Re-Identification Algorithm[J]. Laser & Optoelectronics Progress, 2022, 59(22): 2220001

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享