See our blog post for detailed discussion + results!

The first time:

module load miniconda
conda create -n deepfake python=3.7 pytorch

To install a new package:

conda install opencv

Every time:

./init

in deepfake.

For jupyter notebooks:

./jupinit

then copy the line

MacOS or linux terminal command to create your ssh tunnel
ssh -N -L {port}:{node}:{port} {un}@farnam.hpc.yale.edu

For a GPU-enabled server:

./gpuinit

The training/testing data are in /gpfs/ysm/project/amth/amth_jch97/deepfake/. See https://docs.ycrc.yale.edu/ for more info.

Models: models.py Training: train_*.py Testing: test_*.py Dataset classes: datasets.py Models: *.py Dataset: train/ (50/50 real/fake training set in train/balanced/ and test in train/test_balanced Cache of images already processed through CAE: encode_cache/ or face_encode_cache/