HackMD
At first, we need to setup the environment.
Anaconda is too fat for me to setup, so I choose the minimal installation
Environment
- Git
- Python
Here I recommend to install python via MiniConda installation
Go https://docs.conda.io/en/latest/miniconda.html
Download python3.8 for Linux, because the project is supported below 3.8.
After installation, re-open your terminal.
You will get the base of conda environment.
(base) user@user-machine:~$
Repo
git clone https://github.com/deepfakes/faceswap.git
And it's easy to setup the project
python setup.py
You will get the notification about some of options
- The image file may be corrupted
- The server hosting the image is unavailable
- The image path is incorrect
- The image format is not supported
I keep these options disabled, because I want to train model via CPU only.
OK! almost done.
Data Set
Here, we need to prepare the data set for model training.
so first step, Extract
I prepare two sources of video and photo, and put them into the path below respectively
- faceswap/src/example1
- faceswap/src/example2
Command line:
# extract photos
python faceswap.py extract -i ~/faceswap/src/example1 ~/faceswap/faces/example1
# extract videos
python faceswap.py extract -i ~/faceswap/src/ex1.mp4 ~/faceswap/faces/example1
Model
Start to train our model.
python faceswap.py train -A ~/faceswap/faces/example1 -B ~/faceswap/faces/example2 -m ~/faceswap/ex1_ex2_model/ -p
-p option to preview the process of training
You could stop the training process in any time, but recommend that the faces between A and B are clearly showed at preview screen (Means high accuracy)
Result
Converting
Now, we can convert the video to test the model is good or not.
python faceswap.py convert -i faceswap/src/ex1.mp4 -o ~/faceswap/converted/ -m ~/faceswap/ex1_ex2_model/
After this command, you will get a lot of photos under "converted" folder.
To generate video via these photos.
ffmpeg -i video-frame-%0d.png -c:v libx264 -vf "fps=25,format=yuv420p" out.mp4
Record my command:
# because my photos name: Adi_000001.png ~ Adi_000171.png, so use %3d to present
ffmpeg -i Adi_000%03d.png -c:v libx264 -vf "fps=25,format=yuv420p" out.mp4
