BareSkinNet: De‐makeup and De‐lighting via 3D Face Reconstruction

We propose BareSkinNet, a novel method that simultaneously removes makeup and lighting influences from the face image. Our leverages 3D morphable model does not require reference clean image or specified light condition. By combining process of reconstruction, we can easily obtain geometry coarse textures. Using this information, infer normalized texture maps (diffuse, normal, roughness, specular) by an image-translation network. Consequently, reconstructed textures without undesirable information will significantly benefit subsequent processes, such as re-lighting re-makeup. In experiments, show BareSkinNet outperforms state-of-the-art removal methods. addition, our is remarkably helpful in removing to generate consistent high-fidelity maps, which makes it extendable many realistic generation applications. It also automatically build graphic assets images before after with corresponding data. This assist artists accelerating their work, avatar creation.