Identify deepfakes through original and modified XceptionNet
models. Feed face images to the network and receive predictions whether the face is original/real, or the result of deepfake techniques. Built upon tensorflow
and opencv-python
.
It is recommended to use conda
for this project. Below are the script to replicate the project environment.
Steps:
- Create conda environment. below, the env is named
torchit
. - Activate the conda env
- Install
dlib
throughconda-forge
channel. DO NOT use pip fordlib
as it requires numerous cublas, cudnn and cuda-related libraries. - Install other packages through pip
conda create -n torchit python --yes
conda activate torchit
conda install -c conda-forge dlib --yes
pip install torch torchvision facenet-pytorch opencv-python
Install and run jupyterlab
(optionally) through pip as well with:
pip install jupyterlab jupyterlab-lsp # Install jupyterlab and LSP (language server protocol) to enable documentation, and error checkings
jupyter-lab # run jupyterlab
All codes (python, notebook) are available inside src/
.
This python script and its notebooks are licensed under MIT License.