model optimization

average accuracy/%

training time/min

ResNet

97.0625

24.1

Independent downsampling

97.5938

24.6

inverted residual

98.5938

54.8

DW large convolutional kernel

98.5625

25.1

LN

98.8438

25.3

SE

99.6875

26

GELU

99.7188

25.9