avatarify-fast

One-click video face-swap. This is a pure python scripts without any GUI. It's based on avatarify-python and uses face landmarks lib to implement a pipeline for processing frames automaticly.

I merge code from:

https://github.com/alievk/avatarify-python
https://github.com/AliaksandrSiarohin/motion-cosegmentation
https://github.com/ainrichman/Peppa-Facial-Landmark-PyTorch

How Fast?

12fps!! with TITAN X (Pascal)

Requirements

pytorch
opencv-python==4.2.0.34
face-alignment==1.3.3
pyzmq==20.0.0
msgpack-numpy==0.4.7.1
pyyaml==5.3.1

Weights prepare

Downlowd models' weights of avatarify-python "vox-adv-cpk.pth.tar" from
https://github.com/alievk/avatarify-python/tree/master/docs#download-network-weights

avatarify-python need face-alligment library.

git clone https://github.com/1adrianb/face-alignment
cd face-alignment
pip install -r requirements.txt
python setup.py install

Running Steps

run extract_audio.sh to gain soundtrack mp3 file.
run demo.py with re-write params in local path.
run merge_audio.sh to merge soundtrack and video.

TODO

A better face swap model which cases to less color shift.
Or, maybe a better landmark model which contains brow.
Corner case process, like with cap/hat, large scale head pose, etc.
Maybe we need a face recognition to locate the specific person in multi-persons scene.