Table 1.

Characteristics of the pre-trained CNN architectures adopted in our study.

Network	#Layers	#Learnable Layers	Network Size (MB)	Input Image Size	#Para (Millions)
AlexNet [24]	25	8	227	$227\times 227$	61
Vgg16 [26]	41	16	515	$224\times 224$	138
Vgg19 [26]	47	19	535	$224\times 224$	144
GoogleNet (Inceptionv1) [27]	144	22	27	$224\times 224$	7
Inceptionv3 [28]	315	48	89	$299\times 299$	23.9
ResNet18 [30]	71	18	44	$224\times 224$	11.7
ResNet50 [30]	177	50	96	$224\times 224$	25.6
ResNet101 [30]	347	101	167	$224\times 224$	44.6
InceptionResv2 [29]	824	164	209	$299\times 299$	55.9
Xception [32]	170	71	85	$299\times 299$	22.9
DenseNet201 [31]	708	201	77	$224\times 224$	20
MobileNetv2 [33]	154	53	13	$224\times 224$	3.5
ShuffleNet [35]	172	50	5.4	$224\times 224$	1.4
NasnetMobile [34]	913	*	20	$224\times 224$	5.3
NasnetLarge [34]	1243	*	332	$331\times 331$	88.9
DarkNet19 [36]	64	19	78	$256\times 256$	20.8
DarkNet53 [37]	184	53	155	$256\times 256$	41.6
EfficientNetB0 [38]	290	82	20	$224\times 224$	5.3
SqueezeNet [40]	68	18	5.2	$227\times 227$	1.24