The network receives grayscale input (L channel) and outputs color predictions (*a*b channels). The discriminator uses a PatchGAN design that classifies N×N patches as real or fake, rather than ...