De-Fake

It is the code for the paper: DE-FAKE: Detection and Attribution of Fake Images Generated by Text-to-Image Generation Models.

Environment

You first need to build the environment by:

conda env create -f environment.yaml
conda activate defake

Infer

For the usage, You can download our model on

https://drive.google.com/file/d/1qI7x5iodaCFq0S61LKw4wWjql7cYou_4/view?usp=sharing

and

https://drive.google.com/file/d/1SuenxJP10VwArC6zW0SHMUGObMRqQhBD/view?usp=sharing

for the encoder and classifier.

Then test on

python test.py --image_path XXX

Train

If you want to train the detector yourself, please enter the correct file path in train.py.

Then

python train.py

License

DeFake is licensed under the term of thr MIT license. See LICENSE for more details.