If you wonder how to train using vast.ai or any cloud servers with latest faceswap or if you run across issues with r1.0 edition on cloud servers, then you are like me.
After much of research and reading, below is what got me a successful cloud training server. The configuration is for Ubuntu servers with >15Gb disk space and at least 1 NVIDIA GPU.
Step 1 : Once you ssh into the server, do an update
Code: Select all
apt-get update
Step 2: Install wget and other utils. this will help you with anaconda installation
Code: Select all
apt-get install -y wget git nano p7zip-full
Step 3 -
Code: Select all
wget https://repo.anaconda.com/archive/Anaconda3-2020.11-Linux-x86_64.sh
You can get the latest release anaconda for linux from the anaconda website. Just put the download link to the .sh file in above after wget. https://www.anaconda.com/products/individual#linux
Ste 4 - install anaconda using the step 3 downloaded sh file. Enter yes and check installation path if you would like to change.
Code: Select all
bash <fileName>.sh
Step 5 - Initialize conda virtual environment management using
Code: Select all
source ~/.bashrc
Step 6 - Create your faceswap environment. Use any name of your choice
python version is important for the latest release. I recommend 3.7
Code: Select all
conda create --name faceswap python==3.7
Activate faceswap
Code: Select all
conda activate fswp
Step 7 - install 2.2.0 version of tensorflow gpu. this is most important
Code: Select all
conda install tensorflow-gpu==2.2.0
Stepp 8 - install other libraries. Might need for ui and few code components. better to install
Code: Select all
apt-get install -y libgl1-mesa-glx libegl1-mesa libxrandr2 libxrandr2 libxss1 libxcursor1 libxcomposite1 libasound2 libxi6 libxtst6
Step 9 - Clone faceswap from repo
Code: Select all
git clone --depth 1 https://github.com/deepfakes/faceswap.git
And then finally,
Code: Select all
python faceswap.py
You are good to go once you set the training and other options on config.
PS:
If you land up in error with a clod server coming up with a pre-existing tensorflow, then ensure version is 2.2.0.
Else uninstall the existing and install 2.2.0 and repeat from step 6
pip uninstall tensorflow-gpu
pip uninstall keras
pip install tensorflow-gpu==2.2.0
pip install keras