Table 2.

Detailed architectures of SE-ResNet-50 in the fine-tuning phase. Convolution blocks (SE-ResNet-50 uses the SE-ResNet block [21]) are shown in brackets, with the numbers of blocks stacked.

	Output Size	Kernel	Repeat
Stage 1 (Freeze)	$112\times 112$	convolution $7\times 7$	1
Stage 2 (Freeze)	$56\times 56$	convolution block	3
Stage 3 (Freeze)	$28\times 28$	convolution block	4
Stage 4	$14\times 14$	convolution block	6
Stage 5	$7\times 7$	convolution block	3
	$1\times 1$	global average pooling
Fully connected	$[2048\times 512]$ Fully connected
Fully connected	$[512\times 8]$ Fully connected
Output layer	8-d softmax