Table 2.

Comparisons of different models in our study

Models	Params	FLOPs	Resolution (C*H*W)	Batch size	Epochs	Optimizer	lr
ResNet-50	26 M	4112 M	1*244*244	150	100	Adam	5.0e − 05
EfficientNet-b5	30 M	2413 M	1*244*244	130*	100	Adam	5.0e − 05
CoAtNet-0-rw	27 M	4215 M	1*244*244	150	100	Adam	5.0e − 05

FLOPs floating-point operations, lr learning rate

^*Reduced batch size owing to the GPU memory limitation