Table 3.

Performance of CNN-BPNN predicting facial landmarks and its morphology applied from five CNN models.

Evaluate metrics		ConvNet	InceptionV3	MobileNet	DenseNet	VGG19
Training and validation stage	R 2	0.582	0.836	0.957	0.884	0.919
	MAE (mm)	2.323	1.529	0.176	0.631	0.352
Testing stage	R 2	0.331	0.749	0.934	0.912	0.871
	MAE (mm)	3.412	1.738	0.281	0.442	0.597