Table 1.

Properties of 16 pre-trained CNNs.

CNN	Depth	Size (MB)	Parameters (millions)	Input image size
AlexNet	8	227	61.0	227 $\times$ 227
GoogLeNet	22	27	7.0	224 $\times$ 224
SqueezeNet	18	4.6	1.24	227 $\times$ 227
ShuffleNet	50	6.3	1.4	224 $\times$ 224
ResNet-18	18	44	11.7	224 $\times$ 224
ResNet-50	50	96	25.6	224 $\times$ 224
ResNet-101	101	167	44.6	224 $\times$ 224
Xception	71	85	22.9	299 $\times$ 299
Inception-v3	48	89	23.9	299 $\times$ 299
Inception-ResNet-v2	164	209	55.9	299 $\times$ 299
VGG-16	16	515	138	224 $\times$ 224
VGG-19	19	535	144	224 $\times$ 224
DenseNet-201	201	77	20.0	224 $\times$ 224
MobileNet-v2	53	13	3.5	224 $\times$ 224
NasNet-Mobile	*	20	5.3	224 $\times$ 224
NasNet-Large	*	360	88.9	331 $\times$ 331

*indicates NASNet-Mobile and NasNetLarge networks do not consist of a linear sequence of modules.