Table 3.

Optuna hyperparameters space used for the architecture optimization process.

CNN models			Other parameters
Model name	Variation	# of parameters	Feature extraction	# of filters	16 - 1024 (log)
EfficientNet	B0	5.3M		Pre-trained weights	Yes or No
MobileNet	v3 Small	2.5M	Classification head	# of filters	16 - 1024 (log)
MobileNet	v3 Large	5.5M		# of layers	2 - 12 (step of 2)
RegNet	Y 400mf	4.3M	Training	Learning rate	${10}^{-5}$-${10}^{-2}$(log)
RegNet	Y 800mf	6.4M		Batch size	16
RegNet	X 400mf	5.5M		Augmentation probability	0.3 - 0.7 (step of 0.1)
RegNet	X 800mf	7.3M
ResNet	18	11.7M
ShuffleNet	v2 x1.5	3.5M
ShuffleNet	v2 x2.0	7.4M