Table 4.

Comparison results of accuracy of different algorithms on each data set.

Algorithm	300W-LP		WFLW		AFLW2000-3D
	AR (%)	MR (%)	AR (%)	MR (%)	AR (%)	MR (%)
SqueezeNet [47]	95.61	4.39	96.48	3.52	96.83	3.17
Xception [48]	93.45	6.55	95.23	4.77	95.37	4.63
LBP [49]	95.68	4.32	96.37	3.63	96.51	3.49
ShuffleNetV2 [50]	94.52	5.48	95.42	4.58	94.37	5.63
MobileNetV2 [51]	97.28	2.72	97.48	2.52	94.59	5.41
This method	98.27	1.73	98.36	1.64	97.48	2.52