Optical Music Recognition
PyClef detects musical symbols in score images and applies MIRP, a staff-referenced inference method, to reconstruct note pitch and export reviewable outputs.
pip install pyclef
Overview
PyClef treats detection as one stage of a larger reconstruction pipeline. The detector locates symbols; MIRP uses the detected clef and staff geometry to infer pitch from relative vertical displacement.
The current model uses YOLOv11S trained on 208 DeepScoresV2 Dense classes, with 0.4126 mAP@[0.5:0.95] in the reported evaluation.
The Staff-Referenced Musical Inference Method defines a reference note from the clef and staff, then infers the pitch of detected notes.
Workflow
Load a PDF or image and detect staves, clefs, notes, rests and related notation symbols.
Use MIRP to convert staff-relative geometry into note names and octaves.
Create annotated pages so the recognition result can be inspected visually.
Write MIDI, MP3 and synchronized video files when those outputs are selected.
Desktop application
Select a score, choose the desired outputs, set the BPM and run the pipeline. Results are written to a folder named after the input file.
