Table 1.

Configuration of the deep convolutional network.

Layer type, feature map			Filters	Kernel	Stride	Padding		Learning rate
Input
	224×224×3		—a	—	—	—		0.0 (freeze)
Convolutional
	112×112×64		64	7×7×3	2	3		0.0 (freeze)
Batch normalization
	112×112×64		—	—	—	—		0.0 (freeze)
Max pooling
	56×56×64		1	3×3	2	1		0.0 (freeze)
Layer 1
	Basic block 1-1
		56×56×64	64	3×3×64	1	1		0.0 (freeze)
		56×56×64	64	3×3×64	1	1		0.0 (freeze)
	Basic block 1-2
		56×56×64	64	3×3×64	1	1		0.0 (freeze)
		56×56×64	64	3×3×64	1	1		0.0 (freeze)
Layer 2
	Basic block 2-1
		28×28×128	128	3×3×64	2	1		1e-10
		28×28×128	128	3×3×128	1	1		1e-10
		28×28×128	128	1×1×64	2	0		1e-10
	Basic block 2-2
		28×28×128	128	3×3×128	1	1		1e-10
		28×28×128	128	3×3×128	1	1		1e-10
Layer 3
	Basic block 3-1
		14×14×256	256	3×3×128	2	1		1e-8
		14×14×256	256	3×3×256	1	1		1e-8
		14×14×256	256	1×1×128	2	0		1e-8
	Basic block 3-2
		14×14×256	256	3×3×256	1	1		1e-8
		14×14×256	256	3×3×256	1	1		1e-8
Layer 4
	Basic block 4-1
		7×7×512	512	3×3×256	2	1		1e-6
		7×7×512	512	3×3×512	1	1		1e-6
		7×7×512	512	1×1×64	2	0		1e-6
	Basic block 4-2
		7×7×512	512	3×3×512	1	1		1e-6
		7×7×512	512	3×3×512	1	1		1e-6
Average pooling
	1×1×512		1	7×7	7	0		—
Fully connected layer
	1×7		—	—	—	—		1e-5
Softmax
	1×7		—	—	—	—		—

^aNot applicable.