Figure 2.

Our pix2pix-CGAN architecture. Encoder blocks $CEk$ are from first block $256\times 256\times N_i$ to bottleneck block $1\times 1\times 512$ of Generator. Decoder blocks $CDk$ are from bottleneck block $1\times 1\times 512$ to last block $256\times 256\times N_o$ of Generator. $N_i$, $N_o$ denote the number of input and output image channels of Generator. Discriminator blocks $Ck$ are from block $256\times 256\times (N_i+N_o)$ to block $16\times 16\times 1$. Because of the skip connection, the total channels that pass to Discriminator is $N_i+N_o$. Best viewed in color.