This is the release code repository for the evaluation of PassGAN(https://arxiv.org/pdf/1709.00440.pdf) built using Tensorflow 2, Python 3.7, Keras and Numpy to the described specification. It contains my Tensorflow 2 implementation of an Improved Wasserstein GAN (IWGAN) with the intent of comparing the results found in the aformentioned paper. GCP has the fastest cold start time roughly taking 10minutes from start to train depending on the dataset download speed. Simply grab the container in a VM, install Python 3.7, the latest pip (20 and above) and install requirements. You should be ready to go. Training time on an 80% dataset takes almost a week on a V100, but is characteristically IWGAN stable.
python GAN.py -dataset rock_you -batch_size 64 -layer_dim 128 && tensorboard --log_dir logs/gradient_tape
System Software
Ubuntu 19.10+ or suitable Docker environment https://www.docker.com/get-started
TENSORFLOW-GPU 2.1 https://www.tensorflow.org/
Jetbrains-Pycharm or equivalent
The easiest way to get started
Setup environment according to the TENSORFLOW setup document included: Ubuntu required
Pull the repository
Import packages via Pycharm packet manager
Run project with python GAN.py to pull dataset and begin training