An image translation algorithm based on Generative Adversarial Networks

Image translation is a hot research topic in the field of computer vision. Unlike the traditional method, which may lead to blur and distortion, we aim to investigate a general image-to-image translation algorithm based on the generative adversarial networks in this paper. We introduce the Patch GAN to make sure the image content is high quality in local details. The whole network also learns a loss function to realize the image reconstruction from label maps and edge maps. Through extensive experiments, we demonstrate that the translation effect can be improved by improving the clarity of images.