A CLI for training generative adversarial networks on prespecified artisitic datasets and displaying the results. Written during an exploratory research project during the fall of 2019 with Adam Davenport.
The main program is gans.py, and has two subcommands. Run python3 gans.py train -h and python3 gans.py results -h to see the possible
options. Below is an example of running a training session:
./gans.py train --image-size 64 --nfeatures 32 --learning-rate 0.0001 --batch-size 64 --iteration 50000 --sample-interval 200 --gp-enabled true ross ross-3
One possible way of visualizing the results of the training session is via a GIF, such as the ones below. They show the generator's output for the same latent vectors at regular intervals during the training session.
