TABLE 3.

The accuracy of ResNet models for parts and regions identification based on three levels of data sets.

Data	Code	Type	Epoch	Loss value	Accuracy
					Train (%)	Test (%)	External validation (%)
Parts	Resnet-L	Synchronous	29	0.091	100	100	100
		Asynchronous	49	0.102	100	64	100
		Asys	49	0.219	100	57	100
	Resnet-M	Synchronous	29	0.012	100	100	100
		Asynchronous	45	0.021	100	88	87.5
		Asys	49	0.021	100	96	100
	Resnet-H	Synchronous	29	0.009	100	100	100
		Asynchronous	49	0.027	100	89	90
		Asys	49	0.017	100	81	88
Regions	Resnet-L	Synchronous	29	0.114	100	100	100
		Asynchronous	47	0.248	100	50	25
		Asys	69	0.132	100	54	37.5
	Resnet-M	Synchronous	29	0.030	100	100	100
		Asynchronous	49	0.088	100	62	56.4
		Asys	69	0.045	100	61	66.7
	Resnet-H	Synchronous	27	0.009	100	100	100
		Asynchronous	48	0.011	100	63	62.7
		Asys	47	0.020	100	55	64

Note: The bold value are the optimal results of models under the certain data set.