Table 3. Classification performance of the CNN learning machines in the validation set.

Rows show the performance of the learning machines (general, specific and transfer) on each image view (head, dorsum and profile). Columns show accuracy, average precision and minimum precision performance for each label on top lists. H = head view; D = dorsal view; P = profile view; G = general classifier; S = specific classifier; T = transfer classifier.

		Accuracy			Average precision			Minimum precision
		Top 1	Top 3	Top 5	Top 1	Top 3	Top 5	Top 1	Top 3	Top 5
H	G	0.78	0.91	0.94	0.73	0.90	0.94	0.46	0.81	0.88
	S	0.78	0.90	0.94	0.72	0.90	0.94	0.33	0.83	0.88
	T	0.78	0.91	0.94	0.74	0.91	0.94	0.28	0.83	0.90
D	G	0.62	0.80	0.86	0.54	0.79	0.86	0.15	0.58	0.76
	S	0.56	0.75	0.82	0.50	0.74	0.82	0.14	0.57	0.69
	T	0.60	0.79	0.86	0.54	0.78	0.86	0.11	0.55	0.74
P	G	0.66	0.83	0.89	0.57	0.81	0.88	0.19	0.64	0.77
	S	0.61	0.79	0.85	0.50	0.77	0.84	0.12	0.57	0.72
	T	0.67	0.84	0.89	0.58	0.83	0.88	0.23	0.65	0.80