Table 1.

A List of all layers in the CNN implementation. The thin horizontal line separates the encoder and decoder network parts, but the actual network is not in any way split up.

Input	Layer	Kernel	Stride	Output
256 × 256 × 6	Convolution	3	2	128 × 128 × 32
128 × 128 × 32	Inverted Residual	3	-	64 × 64 × 96
64 × 64 × 96	Inverted Residual	3	-	32 × 32 × 144
32 × 32 × 144	Inverted Residual	3	-	16 × 16 × 192
16 × 16 × 192	Inverted Residual	3	-	8 × 8 × 576
8 × 8 × 576	Convolution	3	2	8 × 8 × 512
8 × 8 × 512	Transposed Convolution	4	1	8 × 8 × 512
8 × 8 × 512	Transposed Convolution	4	2	16 × 16 × 256
16 × 16 × 256	Transposed Convolution	4	1	16 × 16 × 256
16 × 16 × 256	Transposed Convolution	4	1	16 × 16 × 256
16 × 16 × 256	Transposed Convolution	4	2	32 × 32 × 128
32 × 32 × 128	Transposed Convolution	4	1	32 × 32 × 128
32 × 32 × 128	Transposed Convolution	4	1	32 × 32 × 128
32 × 32 × 128	Transposed Convolution	4	2	64 × 64 × 64
64 × 64 × 64	Transposed Convolution	4	1	64 × 64 × 64
64 × 64 × 64	Transposed Convolution	4	1	64 × 64 × 64
64 × 64 × 64	Transposed Convolution	4	2	128 × 128 × 32
128 × 128 × 32	Transposed Convolution	4	1	128 × 128 × 32
128 × 128 × 32	Transposed Convolution	4	2	256 × 256 × 16
256 × 256 × 16	Transposed Convolution	4	1	256 × 256 × 16
256 × 256 × 16	Transposed Convolution	4	1	256 × 256 × 3
256 × 256 × 3	Transposed Convolution	4	1	256 × 256 × 3
256 × 256 × 3	Transposed Convolution	4	1	256 × 256 × 3