PRAIG/JAZZMUS

Optical Music Recognition of Jazz Lead Sheets

🖥️ Slides | 🎶 Poster | Available soon

Dataset

How to use

Check the following code:

import ast

from datasets import load_dataset
from PIL import ImageDraw


DATASET_NAME = "PRAIG/JAZZMUS"

ds = load_dataset(DATASET_NAME)

image = ds["train"][0]["image"]

# list of systems, with bounding boxes and encoding
systems = ast.literal_eval(ds["train"][0]["annotation"])["systems"]

# full page encodings
encoding = ast.literal_eval(ds["train"][0]["annotation"])["encodings"]
mxml_encoding = encoding["musicxml"]
kern_encoding = encoding["**kern"]

# draw the bounding boxes on the image
ImageDraw = ImageDraw.Draw(image)

for idx, s in enumerate(systems):
    print(f"System {idx + 1}:")
    print(f"\t{s['bounding_box']}")
    print(f"\t{repr(s['**kern'])}\n")
    bbox = s["bounding_box"]
    ImageDraw.rectangle(
        [bbox["fromX"], bbox["fromY"], bbox["toX"], bbox["toY"]],
        outline="red",
        width=2,
    )

image.save("image_with_bboxes.pdf", "PDF")

Other dataset versions

🤗 JAZZMUS dataset collection

Citation

We are waiting for the official ISMIR 2025 proceedings.

@misc{martinezsevilla2025omrjazz,
      title={Optical Music Recognition of Jazz Lead Sheets}, 
      author={Juan Carlos Martinez-Sevilla and Francesco Foscarin and Patricia Garcia-Iasci and David Rizo and Jorge Calvo-Zaragoza and Gerhard Widmer},
      year={2025},
      eprint={2509.05329},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2509.05329}, 
}