Table 2.

Architecture of the domain classifier.

	Representation size	Layer
Input	32 $\times$ 32
Conv2D 1	32 $\times$ 32	3 $\times$ 3 kernel, 1024, stride 1, pad 1
ReLU 2	32 $\times$ 32
Conv2D 2	32 $\times$ 32	3 $\times$ 3 kernel, 1, stride 1, pad 1
Sigmoid	32 $\times$ 32
Output	32 $\times$ 32