liebharc/homr

What it solves

homr is an Optical Music Recognition (OMR) tool that converts photos of sheet music into machine-readable MusicXML files. This allows users to import their physical sheet music into digital notation software like MuseScore for editing or playback.

How it works

The software uses a four-stage pipeline to process images:

1. Segmentation: UNet-based models identify structural components like staff lines, note heads, stems, rests, bar lines, and clefs.
2. Staff Detection: The system identifies "staff anchors" and calculates unit sizes to reconstruct staff lines, handling perspective variations and merging related staffs (e.g., grand staffs for piano).
3. Symbol Recognition: Dewarped staff images are processed by a transformer-based model that performs end-to-end sequence recognition to determine rhythm, pitch, and articulation marks.
4. Export: The recognized symbols are are converted into a MusicXML file.

Who it’s for

Musicians, composers, and archivists who need to digitize physical sheet music into a format compatible with standard music notation software.

Highlights

• End-to-end pipeline: Combines segmentation and transformer-based recognition for high accuracy.
• Perspective correction: Includes dewarping to handle camera photos of sheet music.
• Multimodal support: Handles grand staffs and multiple voices on a single staff.
• MusicXML output: Produces industry-standard files compatible with tools like MuseScore.