High-Precision and Lightweight Facial Landmark Detection Algorithm

Lihuai Xu; Zhe Li; Jiajia Jiang; Fajie Duan; Xiao Fu

doi:10.3788/LOP57.241026

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 24 >Page 241026 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 24, 241026 (2020)

High-Precision and Lightweight Facial Landmark Detection Algorithm

Lihuai Xu¹, Zhe Li², Jiajia Jiang^1、*, Fajie Duan¹, and Xiao Fu¹

Author Affiliations

¹State Key Laboratory of Precision Measuring Technology and Instruments, Tianjin University, Tianjin 300072, China

²Institute of Deep-Sea Science and Engineering, Chinese Academy of Sciences, Sanya, Hainan 572000, China

show less

DOI: 10.3788/LOP57.241026 Cite this Article Set citation alerts

Lihuai Xu, Zhe Li, Jiajia Jiang, Fajie Duan, Xiao Fu. High-Precision and Lightweight Facial Landmark Detection Algorithm[J]. Laser & Optoelectronics Progress, 2020, 57(24): 241026 Copy Citation Text

show less

Fig. 1. Overall framework of our algorithm. (a) Teacher network; (b) student network

Download full size

Fig. 2. Basic module of the encoder. (a) Bottleneck; (b) lightweight Bottleneck

Download full size

Fig. 3. Heatmap of key points of real image

Download full size

Fig. 4. Effect image of data augmentation. (a) Horizontal mirror; (b) scale; (c) rotation

Download full size

Fig. 5. Visualization results of different datasets. (a) 300W dataset; (b) WFLW dataset

Download full size

Method	Common	Challenging	Full
RCPR	6.18	17.26	8.35
DSRN	4.12	9.68	5.21
DVLN	3.79	7.15	4.45
PCD-CNN	3.67	7.62	4.44
CPM	3.39	8.14	4.36
SAN	3.34	6.60	3.98
LAB	2.98	5.19	3.49
Teacher(our)	3.03	5.26	3.51
Student(our)	3.35	5.78	3.79
Student+KD (our)	3.21	5.41	3.60

Table 1. NME of different algorithms in 300W dataset unit:%

Method	Test	Pose	Expression	Illumination	Make-up	Occlusion	Blur
CFSS	9.07	21.36	10.09	8.30	8.74	11.76	9.96
DVLN	6.08	11.54	6.78	5.73	5.98	7.33	6.88
MANet	5.41	10.03	5.56	5.54	6.03	7.00	6.25
LAB	5.27	10.24	5.51	5.23	5.15	6.79	6.32
Teacher(our)	5.20	9.01	5.36	5.27	5.45	6.37	6.22
Student(our)	6.04	11.52	6.76	5.70	5.94	7.30	6.87
Student+KD (our)	5.50	9.83	5.89	5.40	5.48	6.88	6.81

Table 2. NME of different algorithms in WFLW dataset unit:%

Method	Backbone	Params /M	FPS/frame
LAB	Houglass	25.1	6
Teacher(our)	ResNet50	25	126
Student+KD (our)	Light-ResNet50	2.8	162

Table 3. Network structure, parameter amount and speed of different algorithms

Method	NME
Student	3.79
Student+Pixel	3.67
Student+Pair	3.62
Student+Pixel+Pair	3.60

Table 4. Accuracy of different loss functions on 300W datasetunit:%

Lihuai Xu, Zhe Li, Jiajia Jiang, Fajie Duan, Xiao Fu. High-Precision and Lightweight Facial Landmark Detection Algorithm[J]. Laser & Optoelectronics Progress, 2020, 57(24): 241026

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information