deepfakesclub
commented
6 years ago If you want to set it up yourself. here is a basic guide.
I find it convenient to copy my own custom portable install to cloud instances, which speeds up creating new instances
This is from WinPython Zero (minimal install version). Install WinPython Zero 3.6 pip install pathlib==1.0.1 pip install scandir==1.6 pip install h5py==2.7.1 pip install Keras==2.1.2 pip install opencv-python==3.3.0.10 pip install tensorflow-gpu==1.5.0 pip install scikit-image
[Slightly tricky part. Install Visual Studio 2015, including C++ tools and SDK, Install CMake, Install boost binaries] git clone https://github.com/davisking/dlib cd dlib python setup.py install --yes USE_AVX_INSTRUCTIONS --yes DLIB_USE_CUDA cd ..
pip install face_recognition (not required for most recent commit, just leaving here for posterity) pip install tqdm
DONE IF YOU ONLY WANT FACESWAP.
git clone https://github.com/dfaker/df cd df git clone https://github.com/keras-team/keras-contrib.git cd keras-contrib python setup.py install cd ..
[Slightly tricky part] the precompiled pytorch wasn't installing cleanly with WinPython Zero for me, so I created a conda install of pytorch as explained here: https://github.com/peterjc123/pytorch-scripts Copy the pytorch folders from the conda environment's site-packages to your WinPython site-package folder.
git clone https://github.com/1adrianb/face-alignment.git cd face-alignment python setup.py install
You now have your own portable python install.
iperov
Clorr
I'm finding that WinPython is the simplest way to share no-install setups, easier than embeddable zip or pyinstaller, etc. WinPython handles all the custom paths for you. These are really easy to setup.
I put together portable python installs of faceswap and dfaker (the 128x128 output repo). That means you just unzip and run the python programs without any installation, compiling dlib, or other annoying things.
https://mega.nz/#!Wcg2mRDC!WnsW2emBHjRry6FkAq7UZjP6VEv_AT5sUfnJIisRtY8
The good thing about WinPython is you can do whatever you want, clone your own repo in. Replicating this is fairly straightforward, but the two possibly tricky parts are compiling dlib with GPU support and installing pytorch (for dfaker).
I tested the basic functions (extract, train, convert with default options), but let me know if something is not working.
Requires CUDA 9, cuDNN 7, VC Redistributable 2015 to be pre-installed, like everything else. Win10, Intel, NVIDIA, you know the drill.
To use:
Unzip the file.
The base directory is pythondf. From here, double click WinPython Command Prompt.exe
Navigate to the directory of the flavor you want to run (within python-3.6.3.amd64). Then, follow the respective github usage notes.
The three included flavors are:
faceswap – Exact copy of most recent faceswap
faceswap_lowmem – The LowMem setting in faceswap doesn’t actually work on 2GB cards. Plus, the new GPU-based face extractor doesn’t work on 2GB cards. For this flavor, I changed these so everything should work on 2GB cards with a batch size of 16 or smaller. (Using Feb 14 commit for GPU face extraction and a different Model_LowMem.py)
dfaker – Exact copy of dfaker with one change. I activated the GPU-based face extractor by commenting one line. You will have to manually change the code if you want to adjust batch size or directory names.