FRGAN: A Blind Face Restoration with Generative Adversarial Networks

Figure 5

First, we input low-quality face image A and distorted face B′ into FaceGAN. The relevant features are extracted with the generator, and face A′ is generated. The prejudgment monitor uses the face information stored in PTNet to distinguish between the face posture and key points. The discriminator identifies the newly generated face A′ and the initial low-quality face A.