Researching | Pedestrian Attribute Recognition Based on Deep Learning

Laser & Optoelectronics Progress, Volume. 57, Issue 6, 061001(2020)

Pedestrian Attribute Recognition Based on Deep Learning

Peipei Yuan and Liang Zhang^*

Author Affiliations

Tianjin Key Laboratory of Advanced Signal and Image Processing, Civil Aviation University of China, Tianjin 300300, China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (9)

References (21)

Paper Information

Figures & Tables(9)

Fig. 1. Pedestrian attribute recognition network

Download full size

Fig. 2. Structure of residual block

Download full size

Fig. 3. Examples of image segmentation. (a) Image to be segmented; (b) result of semantic segmentation; (c) result of instance segmentation

Download full size

Fig. 4. Samples of pedestrian images. (a)-(f) Sample 1-6

Download full size

Fig. 5. Variation trend of regional contrast loss function during training

Download full size

Fig. 6. Result of attribute recognition

Download full size

Table 1. Label of pedestrian image and the ratio of attribute

Table 1. Label of pedestrian image and the ratio of attribute

Attribute	Fig. 4(a)-(c)	Fig. 4(d)-(f)	Ratio/%
Age 16-30	0	0	49.7
Age 31-45	0	1	32.9
Age 46-60	0	0	10.2
Age above 61	1	0	6.2
Backpack	0	0	19.7
Carrying other	0	0	19.9
Casual lower	1	1	86.1
Casual upper	1	1	85.3
Formal lower	0	0	13.8
Formal upper	0	0	13.4
Hat	1	1	10.2
Jacket	0	0	6.9
Jeans	0	0	30.6
Leather shoes	1	0	29.6
Logo	0	0	4.0
Long hair	0	0	23.8
Male	1	0	54.9
Messenger bag	0	1	29.6
Muffler	0	1	8.4
No accessory	0	0	74.9
No carrying	1	0	27.6
Plaid	0	0	2.7
Plastic bag	0	0	7.7
Sandals	0	0	0.2
Shoes	0	1	36.3
Shorts	0	0	3.5
Short sleeve	0	0	14.2
Skirt	0	0	4.6
Sneaker	0	0	21.6
Stripes	0	0	1.7
Sunglasses	0	0	2.9
Trousers	1	1	51.5
T-shirt	0	0	8.4
Upper other	1	1	45.6
V-neck	0	0	1.2

Table 2. Experimental results of PETA dataset%

Table 2. Experimental results of PETA dataset%

Algorithm	E_mA	E_acc	E_prec	E_rec	F1
ACN^[21]	81.15	73.66	84.06	81.26	82.64
DeepMAR^[5]	82.89	75.07	83.68	83.14	83.41
FSPP^[20]	81.67	75.72	84.84	83.10	83.96
HP-Net^[7]	81.77	76.13	84.92	83.24	84.07
PGDM^[18]	82.97	78.08	86.86	84.68	85.76
Experiment1	81.55	76.03	85.20	82.85	84.10
Experiment2	83.37	78.56	87.63	85.03	86.24
Experiment3	84.49	79.44	87.82	85.94	86.87

Table 3. Experimental results of RAP dataset%

Table 3. Experimental results of RAP dataset%

Algorithm	E_mA	E_acc	E_prec	E_rec	F1
ACN^[21]	69.66	62.61	80.12	72.26	75.98
DeepMAR^[5]	73.79	62.02	74.92	76.21	75.56
FSPP^[20]	79.64	60.25	69.10	80.16	74.21
HP-Net^[7]	76.12	65.39	77.33	78.79	78.05
PGDM^[18]	74.31	64.57	78.86	75.90	77.35
Experiment1	79.01	66.85	79.47	80.04	77.94
Experiment2	79.70	67.02	80.54	80.47	78.38
Experiment3	80.11	67.68	81.60	81.55	79.62

Tools

Get Citation

Copy Citation Text

Peipei Yuan, Liang Zhang. Pedestrian Attribute Recognition Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2020, 57(6): 061001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Jun. 26, 2019

Accepted: Aug. 21, 2019

Published Online: Mar. 6, 2020

The Author Email: Liang Zhang (l-zhang@cauc.edu.cn)

DOI:10.3788/LOP57.061001

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology