High fidelity face attribute editing based on TransUNet

In the last few years,with the development of generative adversarial networks (GAN), Significant technical updates have been made in the field of face attribute editing. A new method is proposed in this paper for editing face attributes. Based on the advantages of StyleGAN in face generation, TransUNet and High-Fidelity Encoder are added to the entire network to achieve accurate, controllable and highly realistic editing effects. By integrating the powerful ability of TransUNet to extract image feature information, the structure and semantic information of face images can be accurately captured, and accurate attribute editing can be achieved. In addition, we have designed High-Fidelity Encoder that focuses on maintaining the quality of the visual effects and naturalness of the images during the editing process, minimizing visual artifacts and unnatural appearances, and producing highly realistic editing results that are indistinguishable from real photos. From the results obtained by our experiments, our method has significant advantages in attribute accuracy and visual effect quality.