Table 1.

NI-Module backbone in which NIN indicates the Network-in-Network and IN refers to the Instance Normalization.

	Representation size	Layer
Input	32 $\times$ 32
NIN 1	32 $\times$ 32	1 $\times$ 1 kernel, 1024, stride 1, pad 0
IN 1	32 $\times$ 32	1024 channels
ReLU 1	32 $\times$ 32
NIN 2	32 $\times$ 32	1 $\times$ 1 kernel, 1024, stride 1, pad 0
IN 2	32 $\times$ 32	1024 channels
ReLU 2	32 $\times$ 32
NIN 3	32 $\times$ 32	1 $\times$ 1 kernel, 1024, stride 1, pad 0
IN 3	32 $\times$ 32	1024 channels
ReLU 3	32 $\times$ 32
Output	32 $\times$ 32