In this project, my coursemates and I leveraged high-performing unconditional image generation models (StyleGAN2, Wasserstein GAN & PixelCNN++) to generate music from image representations of MIDI files.
Please see the report writeup for more details.