MagGAN: High-Resolution Face Attribute Editing with Mask-Guided Generative Adversarial Network

We present Mask-guided Generative Adversarial Network (MagGAN) for high-resolution face attribute editing, in which semantic facial masks from a pre-trained parser are used to guide the fine-grained image editing process. With introduction of mask-guided reconstruction loss, MagGAN learns only edit parts that relevant desired changes, while preserving attribute-irrelevant regions (e.g., hat, scarf modification ‘To Bald’). Further, novel conditioning strategy is introduced incorporate influence region each change into generator. In addition, multi-level patch-wise discriminator structure proposed scale our model (\(1024 \times 1024\)) editing. Experiments on CelebA benchmark show method significantly outperforms prior state-of-the-art approaches terms both quality and performance.