face-generator

Face generation with Deep Convolutional Generative Adversarial Networks (DCGAN).

Uses the CelebFaces Attributes Dataset (CelebA) to train the networks, which then produces new faces (which are mostly white people due to bias in training dataset).

Example output:

Setup notes

I'm using AWS' Deep Learning AMI (Ubuntu) v23 (however to run remotely on CPU follow the binder link).

# enter provided env
source activate pytorch_p36
python -m pip install -r requirements.txt

# quick setup jupyter
jupyter notebook --generate-config
sed -ie "s/#c.NotebookApp.ip = 'localhost'/#c.NotebookApp.ip = '*'/g" ~/.jupyter/jupyter_notebook_config.py
jupyter notebook --ip=0.0.0.0 --no-browser

# navigate to url provided (will need to replace IP)
# may also need to select kernal corresponding to env above

Images

Need to download:

• CelebFaces Attributes Dataset (CelebA)
  • Goes in: <project_root>/assets