Image Translation by Domain-Adversarial Training

Figure 3

Different models and several configurations of our feature comparator for image inpainting: (a) input; (b) ground-true; (c) context-encoder [1]; (d) pix2pix [18]; (e) our model with all the ReLU layers of the discriminator; (f) our model with only the last ReLU layer of the discriminator; and (g) our model with all the last ReLU layer of each block of VGG19 [46].

(a)

(b)

(c)

(d)

(e)

(f)

(g)