Skip to content
from deekfakes' faceswap: https://www.reddit.com/user/deepfakes/
Python
Branch: master
Clone or download
joshua-wu Update README.md
add message from deepfakes
Latest commit 00b558f Dec 16, 2017
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
README.md Update README.md Dec 16, 2017
image_augmentation.py first commit Dec 15, 2017
model.py first commit Dec 15, 2017
pixel_shuffler.py first commit Dec 15, 2017
script.py
train.py
training_data.py first commit Dec 15, 2017
umeyama.py first commit Dec 15, 2017
utils.py first commit Dec 15, 2017

README.md

deepfakes_faceswap

This is the code from deepfakes' faceswap project. Hope we can improve it together, HAVE FUN!

Message from deepfakes:

Whole project with training images and trained model (~300MB):
anonfile.com/p7w3m0d5be/face-swap.zip or click here to download

Source code only:
anonfile.com/f6wbmfd2b2/face-swap-code.zip or click here to download

Requirements:

Python 3
Opencv 3
Tensorflow 1.3+(?)
Keras 2

you also need a modern GPU with CUDA support for best performance

How to run:

python train.py

As you can see, the code is embarrassingly simple. I don't think it's worth the trouble to keep it secret from everyone. I believe the community are smart enough to finish the rest of the owl.

If there is any question, welcome to discuss here.

Some tips:

Reuse existing models will train much faster than start from nothing.
If there are not enough training data, start with someone looks similar, then switch the data.

You can’t perform that action at this time.