Deep Realistic Facial Editing via Label-restricted Mask Disentanglement

Figure 5

Overall structure of LMGAN. (a) Single attribute training. (b) Multiple attributes parallel training with global discriminator to keep image overall fidelity.