Large Mask Image Completion with Conditional GAN

Recently, learning-based image completion methods have made encouraging progress on square or irregular masks. The generative adversarial networks (GANs) been able to produce visually realistic and semantically correct results. However, much texture structure information will be lost in the process. If missing part is too large provide useful information, result ambiguity, residual shadow, object confusion. In order complete mask images, we present a novel model using conditional GAN called coarse-to-fine condition (CF CGAN). We use generator with symmetry new perceptual loss based VGG-16. symmetric structure. For completion, our method produces generalization ability of also excellent. evaluate CelebA dataset FID, LPIPS, SSIM as metrics. Experiments demonstrate superior performance terms both quality reality free-form completion.