Mashcima 2

A library and a framework for synthesizing images containing music, intended for use as training data for OMR models.

A tutorial

Note: The tool is under development and does not yet have a fixed public API. The tutorial is being developed and from it the whole tool is being built top-down.

To quickly learn how to start using Mashcima for your project, start with the tutorial.

Mashcima Tutorial

How it works

Mashcima is not really a single tool, but more like a framework. To fully learn how to leverage its abilities, start by reading its documentation:

Mashcima Documentation - Introduction

After cloning

Create a virtual environment and install dependencies:

python3 -m venv .venv
.venv/bin/pip3 install -e .

# to run jupyter notebooks:
.venv/bin/pip3 install -e .[jupyter]

TODO list

The first milestone to hit is to convert Mashcima 1 into the modular structure of Mashcima 2, thereby settling most of its interfaces. The only additional logic is full-page synthesis.

• The Scene data model
  • The data model for the output of the synthesizer
  • Inspired by and closely tied to SVG
  • That can accomodate lots of different annotations (consider COCO compatibility)
  • Physically-based coordinate system (millimeters, DPI)
  • Aims to desribe one page of paper - one physical document
• Page synthesizer
  • creates a page of paper with stafflines
  • very crude to begin with, nothing complicated
• Staff sythesizer
  • the Mashcima 1, just modularized
• Symbol synthesizer
  • the Mashcima 1, just modularized
• Assets layer - AssetBundle as a "library" of assets
  • can download, process, and prepare external datasets to be used for the syntehsis process; this incudes trained ML models
  • Asset = a resource used during synthesis
• Symbolic logic
  • parsing & conversions of formats
  • MusicXML as the primary input format
  • generation of synthetic music

Packaging and development notes

• Read this: https://packaging.python.org/en/latest/tutorials/packaging-projects/
• Package configuration inspired by this: https://github.com/vega/altair/blob/main/pyproject.toml
• For development setup inspiration check out: https://altair-viz.github.io/getting_started/installation.html#development-installation
• jupyter notebooks in git: https://mg.readthedocs.io/git-jupyter.html
• deploying voila: https://voila.readthedocs.io/en/stable/deploy.html