This is a fork of https://github.com/lbfs/DeepFaceLab_Linux.
This fork uses a docker container to install DeeFaceLab_Linux instead of Anaconda3. The changes made regard the installation process only. No changes have been made to DeepFaceLab itself.
Notes:
Currently, there's only one option available to install DeepFaceLab Linux.
That is, you have cloned this repository before you start the installation. The provided docker/Dockerfile
does not download this repository at the moment.
Note: The user running the docker container must be member of the group "docker". This example is for Archlinux. Docker is available in repositories of many other Linux distros.
# pacman -S docker
# usermod -a -G docker ${USER}
# systemctl start docker
$ git clone https://github.com/aspera-non-spernit/DeepFaceLab_Linux
$ cd DeepFaceLab_Linux
$ chmod 700 docker_from_cloned_repository.sh
Note: If you experience network issues during the installation often it's Docker having problems to use the System DNS. You can manually edit the file and enter any DNS of your choice:
$ nano docker/daemon.json
// enter your DNS (example dns)
{
"dns": ["10.0.0.2", "8.8.8.8"]
}
# cp docker/daemon.json /etc/docker
Note: The build script will ask you for the environment you are using. And a few other questions.
$ ./docker_from_cloned_repository.sh
Do you want to run on cuda, cpu or opencl [cuda, cpu, opencl]?:
_
// if you answer opencl
What's your graphics card [mesa, nvidia, nvidia-390xx, ivybridge, haswell]?:
_
Where do you want to have the workspace directory [default: /home/${USER}/DeepFaceLab/workspace)]:
You can run and enter your Container with the command:
docker run -ti -v /home/${USER}/workspace:/app/DeepFaceLab_Linux/workspace aspera_non_spernit/deepfacelab
You can copy your video material into the workspace folder on the host machine.
DeepFaceLab_Docker will pick up changes.
Run DeepFaceLab_Docker now? [y, n]? n
Installation successful. Have fun.
Note: DeepFaceLab suggests to delete untrainable face extractions. Instead this version has, during the installation, created two additional folders in the workspace:
../data_src/aligned_notrain
../data_dst/aligned_notrain
You can copy aligned png images that you do not want to train, but that may be required after the training to render a video sequence without mapped faces. These images can be blurry or otherwise untrainable images. You never know if you need them again. If you do you save time for an additional face extract.
You have two options:
- If you have used the default workspace:
/home/${USER}/DeepFaceLab/workspace
you can use the run.sh script:
$ /home/${USER}/DeepFaceLab/run.sh
- If you have changed the workspace directory you have to pass this info as -v argument
// run the container
$ docker run -ti -v {PATH_TO_YOUR_WORKSPRACE}:/app/DeepFaceLab_Linux/workspace aspera_non_spernit/deepfacelab
Note: The idea is that you execute the docker container with appropriate arguments to run a specific task of DeepFaceLab_Linux without logging into the container.
I am mainly using ArchLinux and tested the installation on my local machine and on a remote sanbox cloud machine by ovh. Cloud machines be may need an update before you can install required packages. Often the keys are missing or outdated.
If you are on ArchLinux host in the cloud, you may have to initialize the keys before the first use of pacman:
# pacman-key --init
# pacman-key --refresh-keys
# pacman -Syu
- There have been no changes made to the code base.
- The Russian and Chinese language documentation in docx and pdf format have been removed. Links to the orihinal source are included in this README.md
- Several files in
doc/*.md
have been condensed into one filedocumentation.md