CollageNet

July 12, 2021 ยท View on GitHub

This is the repository for code and demo of the ISMIR 2021 paper CollageNet. The code is incomplete, and will be completed in the future.

code

Training

  • CollageNet: sh scripts/train_adversarial.sh
  • EC2-VAE: sh scripts/train_EC2.sh
  • poly: sh scripts/train_poly.sh

Dependencies

  1. python 3.8.8

  2. pytorch 1.7.1

  3. cudatoolkit 11.0.221

  4. pretty-midi 0.2.9

Reference

https://github.com/buggyyang/Deep-Music-Analogy-Demos

https://github.com/ZZWaang/polyphonic-chord-texture-disentanglement

https://github.com/music-x-lab/POP909-Dataset

demos

fusion demo

Some fusion demos along with the one in Figure 1 of the paper.

The user control inputs are recorded in the file names.

user control demo

Several outputs of the model with different user control inputs.

The user control input cmel=c_mp=c_mr=1-c_ac=1-c_at