The RenaissanceML GAN is an image abstraction generator based on a data set of Renaissance age images.
The inspiration came from projects like StarGAN and CycleGAN (check those out) which showcased the creative and, quite honestly, mind-blowing part of AI.
We built our project with GitHub libraries that , with some fine tuning, created abstractions.
We faced many challenges - image hosting, clean data, lots of data, and TRAINING. GANs are extremely hard to train as you’re training 2 models at once. The smallest change in hyperparameters could mean glorious success or miserable failure.
Anyway, we had a ton of fun designing it, and are 100% going to continue working on stronger and more lifelike GANs in the future!
Log in or sign up for Devpost to join the conversation.