基于生成对抗网络的图像翻译现状研究

Research the status of image translation based on generative adversarial networks

近年来生成对抗网络在计算机视觉领域快速发展,尤其在图像数据生成领域占据着主导地位。图像翻译是在不改变图像内容的前提下,将图像转换成另一领域的图像,类似语言翻译。研究总结了基于生成对抗网络的图像翻译方法的最新进展,详细介绍并对比了近两年来的各个方法(Pix2Pix、Cycle GAN、Star GAN等)基本原理与网络框架结构,最后,探讨了在相关领域的应用前景、目前存在的问题以及后续发展趋势。

In recent years, the generation of confrontation networks has developed rapidly in the field of computer vision, especially in the field of image data generation. Image translation is the conversion of an image into another field without changing the image content, similar to language translation. This paper reviews the recent developments of image translation methods based on generating anti-networks. It introduces and compares the basic principles and network framework of each method (Pix2Pix, Cycle GAN, Star GAN) in the past two years, and discusses the application prospects in related fields. Current problems and subsequent development trends.