Face Completion Based on Symmetry Awareness with Conditional GAN

Face completion is an important topic in the field of computer vision and image processing. Its core task to restore information, so that generated results are as consistent possible with ground-truth results. In existing methods, there no strong constraint on consistency between result true value, symmetry characteristics face ignored, which makes it impossible generate a natural for any position symmetrical face. response these problems, we propose novel method called generative adversarial network (FC GAN). Our generator uses u-net-like structure, discriminator combination global local discriminator. We use new perceptual loss based VGG-19 constrain value. awareness our takes full advantage features optimize completion. For irregular mask completion, produces visually realistic semantically correct evaluate model CelebA dataset FID SSIM indicators. Compared this paper has certain improvement visual effects evaluation