CNN-generated images are surprisingly easy to spot... for now

TL;DR

This paper investigates whether a single CNN-based detector can reliably distinguish real images from a wide variety of CNN-generated images, revealing shared flaws in current generative models.

Contribution

It demonstrates that a classifier trained on one CNN generator can generalize to detect images from various other architectures and datasets, indicating common systematic flaws.

Findings

A classifier trained on ProGAN generalizes well to other models.

CNN-generated images share systematic flaws that hinder realism.

The approach achieves robust detection across multiple architectures.

Abstract

In this work we ask whether it is possible to create a "universal" detector for telling apart real images from these generated by a CNN, regardless of architecture or dataset used. To test this, we collect a dataset consisting of fake images generated by 11 different CNN-based image generator models, chosen to span the space of commonly used architectures today (ProGAN, StyleGAN, BigGAN, CycleGAN, StarGAN, GauGAN, DeepFakes, cascaded refinement networks, implicit maximum likelihood estimation, second-order attention super-resolution, seeing-in-the-dark). We demonstrate that, with careful pre- and post-processing and data augmentation, a standard image classifier trained on only one specific CNN generator (ProGAN) is able to generalize surprisingly well to unseen architectures, datasets, and training methods (including the just released StyleGAN2). Our findings suggest the intriguing possibility that today's CNN-generated images share some common systematic flaws, preventing them from achieving realistic image synthesis. Code and pre-trained networks are available at https://peterwang512.github.io/CNNDetection/ .