Are GAN generated images easy to detect? A critical analysis of the state-of-the-art

In recent years, there has been intense research on the generation of synthetic media, and a large number of deep learning-based methods have been proposed to this end. Generative adversarial networks (GAN), in particular, have brought tremendous quality improvements. There are GAN-based methods to generate images from scratch as well as to modify the attributes of an existing image. A number of exciting applications exist already. However, this technology can also be used for malicious purposes, for example to generate fake profiles on social network or to generate fake news. Even the most careful observer can now be fooled by GAN-generated images, not to mention the average Internet user. Therefore, there is urgent need for automatic tools that can reliably distinguish real content from manipulated content.