표 1. | Table 1. CNN 모델별 layer의 하이퍼 파라미터 | Hyper parameters in layer by CNN model.

Output size	Basic CNN structure	GoogLeNet	ResNet	DenseNet
50×50	Convolution, 64, 7×7+2(S)
25×25	Max pooling, 3×3+2(S)
	Convolution, 128, 7×7+1(S)	Convolution, 64, 1×1+1(S)	[ 64 , 3 × 3 64 , 3 × 3 ] × 3	[ 16 , 3 × 3 4 , 3 × 3 ] × 6
		Convolution, 192, 3×3+1(S)		Convolution, 88, 1×1+1(S)
13×13	Max pooling, 3×3+2(S)	Max pooling, 3×3+2(S)	[ 128 , 3 × 3 128 , 3 × 3 ] × 4	Avg pooling, 2×2+2(S)
	Convolution, 256, 7×7+1(S)	[ 64 , 128 , 32 , 32 96 , 16 ]		[ 16 , 3 × 3 4 , 3 × 3 ] × 12
		[ 128 , 192 , 96 , 64 128 , 32 ]		Convolution, 136, 1×1+1(S)
7×7	Max pooling, 3×3+2(S)	Max pooling, 3×3+2(S)	[ 256 , 3 × 3 256 , 3 × 3 ] × 6	Avg pooling, 2×2+2(S)
		[ 192 , 208 , 48 , 64 96 , 16 ]
		[ 160 , 224 , 64 , 64 112 , 24 ]		[ 16 , 3 × 3 4 , 3 × 3 ] × 24
	Convolution, 512, 7×7+1(S)	[ 128 , 256 , 64 , 64 128 , 24 ]
		[ 112 , 288 , 64 , 64 144 , 32 ]		Convolution, 232, 1×1+1(S)
		[ 256 , 320 , 128 , 128 160 , 32 ]
4×4	Max pooling, 3×3+2(S)	Max pooling, 3×3+2(S)	[ 512 , 3 × 3 512 , 3 × 3 ] × 3	Avg pooling, 2×2+2(S)
		[ 256 , 320 , 128 , 128 160 , 32 ]		[ 16 , 3 × 3 4 , 3 × 3 ] × 16
		[ 384 , 384 , 128 , 128 , 192 , 48 ]
1×1	Global Avg pooling, fully connected