LAUSR

Face attribute editing is to edit the face image by modifying single or multiple attributes while maintaining the face identity. In the paper, we propose a method for attribute editing of face images by using the generative adversarial networks: conditional generative adversarial nets is used as the backbone of the framework and input attributes as conditions to the generator, the generator combines the encoder–decoder with U-Net, and the attribute classifier is added to guarantee the correct attribute operation on the generated image. The receptive field of a single discriminator is very limited, especially when the size of the training picture becomes larger, which will affect the extraction of information. In this paper, we tackle these limitations by using multi-scale discriminators to guide the generator to generate better details. It can macroscopically grasp the global information of the generated pictures and obtain more information of the receptive field. We demonstrate the effectiveness of our method and generate well-preserved facial detail images on CelebA dataset. The fidelity of the generated image is improved, and the method has better flexibility. The experiments show that our method is effective on the real-world dataset.