An Application Comparison of GAN-based Image Translation Methods

Image translation is an important and challenging area of computer vision. It aims to design models that translate source-domain images to target-domain images with applications such as data enhancement, style migration and super-resolution. Until 2016, researchers have successfully implemented generative adversarial networks to image translation using the findings of deep learning techniques for visual generation tasks. In this paper, we first review and analyze the literature in the field of GAN-based image translation and summarize common normalization techniques and model judging metrics. In accordance with the mapping connection between the model inputs and outputs, self-supervised and unsupervised methods to image translation models are then split into categories. The strengths and weaknesses of these models are then analyzed, and a short explanation of how each model performed on various evaluation indicators. Finally, several potential future research issues in this field are discussed.