This is the reproducing version of FaceShiter
Paper Reference: https://arxiv.org/pdf/1912.13457.pdf
Code Reference: https://github.com/taotaonice/FaceShifter
- Python 3.7
- torch 1.4.0
- torchvison 0.5.0
- opencv-python 4.2.0
- numpy 1.18.1
- Flask
- flask_cors
- requests
You can install these packages by pip
pip install -r requirements.txt
If you want to train the model, you have to install apex, you can download and install from the Nvidia/apex: https://github.com/NVIDIA/apex
- URL for download pre-trained weights of arcface: https://drive.google.com/open?id=15nZSJ2bAT3m-iCBqP3N_9gld5_EGv4kp, then put it to the directory "face_modules/"
- URL for download pre-trained weights of AEI-Net: https://drive.google.com/open?id=1iANX7oJoXCEECNzBEW1xOpac2tDOKeu9, then put it to the directory "saved_models/"
You can swap face of two images. One is source image and another is target image
- open config.txt
- set parameters of "source_image_path", "target_image_path", and "result_image_save_path" in section [image_inference]
- run
python image_inference.py
You can swap the faces in video by one source image and one target video
- open config.txt
- set parameters of "source_image_path", "target_video_path", "target_frames_path", "result_frames_path", "result_video_save_path", and "fps" in section [video_inference]
- run
python video_inference.py
You can train the model
- open config.txt
- set parameters of "dataset_path" in section [train]
- run
python train.py - You can modify parameters of BatchSize, epoch,learning rate in train.py
You can deploy model server.
- open config.txt
- set parameters of "port" in section [faceshifter_server]
- run
python faceshifter_server.py
When you deploy the server, you can test it by python-version client
- open config.txt
- set parameters of "url", "source_image_path", "target_image_path", "result_image_save_path" in section [faceshifter_client]
- run
python faceshifter_client.py