- Create a conda virtual environment and activate it.
conda create --name fakeface python=3.8.13
conda activate fakeface
conda install pytorch==1.13.1 torchvision==0.14.1 -c pytorch
- Install the dependent libraries.
pip install -r requirements.txt
- Download the five data sets respectively, unzip them to the data folder, and use the data processing code in data to process the data sets.
/data
deepfake_celeb
fake/
true/
deepfake_dfmnist
fake/
true/
forgery
fake/
true/
gan_wanghong
fake/
true/
gan_yellow
fake/
true/
- We use one Nvidia Tesla 4090 (24G) GPU for training. You need to modify the
trainval_path
in the config.py
file before training.
- You need to modify the
folder_name
in the evaluate.py
file before testing.