Table 1: Configuration of reconstruction model in VGG 19.

	ConvNet r1	ConvNet r2	ConvNet r3	ConvNet r4	ConvNet r5	ConvNet r6	ConvNet r7
Transpose(...)	(2,2)	(2,2)	(2,2)	(2,2)	(2,2)	(2,2)	–
#x(w,h,d)	(8,8,64)	(16,16,64)	(32,32,64)	(64,64,64)	(128,128,64)	(256,256,64)	–
concat(...)	(input x, ConvNet5)	(input x, ConvNet4)	(input x, ConvNet3)	(input x, ConvNet2)	(input x, ConvNet1)	–	–
#x(w,h,d)	(8,8,64+512)	–	–	–	–	–	–
[weight],[bias]	5×5×64,64	5×5×64,64	5×5×64,64	5×5×64,64	5×5×64,64	5×5×64,64	5×5×3,3
Num. ConvNet	ConvNet r1-4	ConvNet r2-4	ConvNet r3-4	ConvNet r4-2	ConvNet r5-2	ConvNet r6-2	ConvNet r7-1
#x(w,h,d)	(8,8,64)	(16,16,64)	(32,32,64)	(64,64,64)	(128,128,64)	(256,256,64)	(256,256,3)