- train.py contains the training module that is initialized in master before the training sessions begins
- master.py loads the trainer and data to begin training
- gan.py holds the Discriminator and Generator Architectures for the base models that are trained
- config contains hyper-params, path to training files and other notes/conditions to be imposed by training session
Sessions can be run locally or on virtual servers. Colab notebook is still in development and not provided yet.
- Use requirements.txt to create a conda repository.
- Simply set the config.json files to point to data files and results directory. Set hyperparameters and set up training session.
- Go to master.py and run the file.
- Repository includes pre-trained models that can be used to generate samples. Notebook for steps to do this is work in progress, will update soon.
- Repository does not include data because data files are large. Find training data here.
If you have any questions, feel free to contact me at ebalogun[AT]stanford[DOT]edu