GAN Image Detection

This repository contains the best GAN generated image detector, namely <ResNet50 NoDown>, among those presented in the paper:

Are GAN generated images easy to detect? A critical analysis of the state-of-the-art
Diego Gragnaniello, Davide Cozzolino, Francesco Marra, Giovanni Poggi and Luisa Verdoliva.
In IEEE International Conference on Multimedia and Expo (ICME), 2021.

The very same architecture has been trained with images generated either by the Progressive Growing GAN or the StyleGAN2 architecture.
To download the trained weights, run:

wget -e robots=off -nd -P ./weights -A .pth -r https://www.grip.unina.it/download/prog/GANdetection/weights

or manually download them from here and put them in the folder weights.

Requirements

• python>=3.6
• numpy>=1.19.4
• pytorch>=1.6.0
• torchvision>=0.7.0
• pillow>=8.0.1

Test on a folder

To test the network on an image folder and to collect the results in a CSV file, run the following command:

python main.py -m weights/gandetection_resnet50nodown_stylegan2.pth -i ./example_images -o out.csv

Traning-set

1) ProGAN Traning-set

For training using ProGAN images, we used the traning-set provided by "CNN-generated images are surprisingly easy to spot...for now"

2) StyleGAN2 Traning-set

For training using StyleGAN2 images, we generated the fake images and used different public datasets for pristine images. The 720K fake images can be downloaded here. The number of pristine images for each public dataset is reported in the following table:

Dataset	Size	#Images
LSUN cat	256x256	120000
LSUN church	256x256	120000
LSUN hourse	256x256	120000
LSUN car	512x512	120000
AFHQ cat	512x512	4700
AFHQ dog	512x512	4700
AFHQ wild	512x512	4700
AnimalWeb	512x512	14100
BreCaHAD	512x512	1800
FFHQ	1024x1024	28800
MetFaces	1024x1024	14100

We downloaded and processed the pristine images following the guide provided by StyleGAN2-ada. The links to public datasets are reported in the guide except AnimalWeb dataset that can be downloaded here.