A VAE trained on the image representation of midi files.
groove-v1.0.0-midionly.zip from the Groove MIDI Dataset
-
data_preprocessing.py: converts midi to images using https://github.com/mathigatti/midi2img. -
train.py: train the vae -
analysis.py: run analysis on trained model
I am still exploring this idea. Feel free to clone this repo and contribute if you like it.
I am sharing a model and some examples in the generations/ folder.
The code is mostly based on the VAE series by Valerio Velardo.