Query-and-reArrange

Repository for Paper: Zhao et al., Q&A: Query-Based Representation Learning for Multi-Track Symbolic Music re-Arrangement, in IJCAI 2023 Special Track for AI the Arts and Creativity.

Demo page

https://zhaojw1998.github.io/Query_and_reArrange

Code and File Directory

This repository is organized as follows:

root
  ├──checkpoints/         model checkpoints
  │    
  ├──data/                processed data and pre-processing scripts
  │    
  ├──demo/                demo save directory
  │       
  ├──dl_modules/          Q&A model's sub-modules
  │    
  ├──utils/               scripts for utility functions
  │    
  ├──dataset.py           dataset and loader
  │   
  ├──model.py             Q&A model
  │   
  ├──train.py             traning script
  │ 
  └──inference.ipynb      tutorial for running the model

How to run

• Q&A is now on Google Colab, where you can quickly test our model online.
• Alternatively, follow the guidance in ./inference.ipynb offline for more in-depth testing.
• If you wish to train our model from scratch, run ./train.py. You may wish to configure a few params such as BATCH_SIZE from the beginning of the script. When DEBUG_MODE=1, it will load a small portion of data and quickly run through for debugging purpose.
• Dependencies of our work includes pytorch (ver. >= 1.10), pretty_midi, scipy, tensorboard, and tqdm.

Data

• For details about the data we use, please refere to ./data.

Contact

Jingwei Zhao (PhD student in Data Science at NUS)

[email protected]

June. 04, 2022