Face Translation based on Semantic Style Transfer and Rendering from One Single Image
Peizhen Lin, Baoyu Liu, Lei Wang, Zetong Lei, Jun Sheng Cheng · 2021
Many avatar characters have been animated in films or games, which always need a lot of time for post-processing with the computer graphics technologies. In recent years, lots of deep learning based methods have been proposed for face translation and image generation, which always require a large amount of data for training. However, there are few samples for special characters' prototype. In this paper, we present one face translation framework for translating human faces to that with visual effects from one single prototype image. The proposed framework consists of three modules. We first design one module to generate semantic face mask–the semantic mask generating (SMG) module. According to the semantic mask, the face color tone can be changed to that of the prototype. So we design the semantic color transfer (SCT) module. For the local textures, we design the deformation and rendering (DR) module. Experiments show that the proposed framework can generate images with prototype's visual effects while preserving the original person's identification and expression information.