A Dataset and Benchmark for Optical Music Recognition of String Quartet Scores

This is the umbrella repository for the paper:

A Dataset and Benchmark for Optical Music Recognition of String Quartet Scores
Dongmin Kim, Brian Liu, Jose J. Valero-Mas, Dasaem Jeong
Proceedings of the 27th International Society for Music Information Retrieval Conference (ISMIR), 2026

This release introduces OSSQ-OMR, a dataset of string quartet movement scores with paired synthetic and scanned images and multiple symbolic-music target formats, together with a benchmark suite that evaluates several optical music recognition (OMR) architectures — including a hybrid SMT/Zeus encoder-decoder family and the LEGATO baseline — on transcribing these scores, measured with the OMR-NED metric. This repository is the entry point that explains what was released and links out to the four component repositories that hold the actual code and data history.

What's in this release

Repo	Role	License
MALerLab/ossq-omr	the OSSQ-OMR dataset — MuseScore/MusicXML annotation sources and their revision history; bulk derived artifacts distributed via Zenodo and Hugging Face	CC0 1.0 (data)
MALerLab/omr-data-preprocessor	the preprocessing pipeline that builds every derived symbolic format from the dataset sources	MIT
MALerLab/sqomr	model training and evaluation experiments	MIT
MALerLab/lmxe	the LMXE symbolic format library (derived from OMR-Research/lmx, © 2024 Jiří Mayer)	MIT

How the pieces fit together

flowchart LR
    subgraph Sources
        A["ossq-omr\n(OSSQ-OMR dataset repo)\nMuseScore annotation history"]
    end

    subgraph Pipeline
        B["omr-data-preprocessor\n(omrdp)"]
        L["lmxe\n(LMXE format library)"]
    end

    subgraph Derived["Derived formats"]
        D["MusicXML / LMXE / PLMXE / RLMXE\nKRN / EKRN / ABC / EABC\n(aligned, segmented, rendered images)"]
    end

    subgraph Bulk["Bulk artifact channels"]
        Z["Zenodo (DOI)"]
        H["Hugging Face"]
    end

    subgraph Benchmark
        S["sqomr\ntraining / evaluation"]
        LG["LEGATO baseline\n(submodule)"]
    end

    R["Results\n(OMR-NED)"]

    A -- "raw scores" --> B
    L -. "format spec/lib" .-> B
    B --> D
    D --> Z
    D --> H
    Z --> S
    H --> S
    S --> LG
    S --> R

Loading

Plain-text summary: the dataset repo holds the source-of-truth MuseScore edit history; the omrdp pipeline (using the lmxe library) converts that into derived symbolic formats and rendered/scanned images, which are distributed in bulk via Zenodo and Hugging Face; sqomr consumes those bulk artifacts (plus the LEGATO baseline as a submodule) to train and evaluate OMR models and produce the benchmark results.

Getting started

Pick the path that matches what you want to do.

(a) I just want the data

Download the pre-built dataset from Zenodo (versioned, DOI'd) or Hugging Face (convenient for streaming/datasets loading):

• Zenodo: (not yet published)
• Hugging Face: (not yet published)

If you also want the full, auditable per-score annotation history (not just the bulk export), see ossq-omr.

(b) I want to reproduce the benchmark

1. Clone sqomr recursively (it includes the LEGATO baseline as a submodule):

   git clone --recursive https://github.com/MALerLab/sqomr.git

2. Obtain the dataset via Zenodo/Hugging Face (see (a) above).
3. Follow the training/evaluation instructions in the sqomr README — command-level details live there rather than being duplicated in this umbrella README.

(c) I want to rebuild the dataset from source

1. Clone ossq-omr to get the raw MuseScore annotation history.
2. Clone omr-data-preprocessor and follow its pipeline steps to convert MuseScore → MusicXML → LMXE/PLMXE/RLMXE/KRN/EKRN/ABC/EABC, then align, segment, and render.
3. omrdp depends on lmxe for the LMXE format; install/reference it as instructed in the omrdp README.

Dataset at a glance

OSSQ-OMR consists of string quartet movements with:

• Both synthetic (rendered) and scanned score images.
• Scores segmented both systemwise and partwise.
• Symbolic transcription targets in several families: LMXE (LMXE / PLMXE part-by-part / RLMXE reduced), KRN (KRN / EKRN), and ABC (ABC / EABC).
• 116 string quartet works by 47 composers, encoded as 122 MuseScore files (two multi-movement works are split by movement), published between 1773 and 1994.
• 24,544 system images and 98,172 staff (part) images: 13,240 synthetic / 11,304 scanned at the system level, and 52,960 synthetic / 45,212 scanned at the staff level.
• Source types across the corpus: 93 full scores, 17 part-books and 6 manuscripts, spanning 32 publishers; the scanned track covers the 93 full scores.
• Splits are drawn at the score level to prevent cross-split leakage, as four random splits with mutually exclusive test sets.

Benchmark at a glance

Architectures evaluated:

• SMT
• Zeus
• Hybrid smt-zeus and zeus-smt encoder-decoder combinations
• LEGATO baseline

Evaluation metric: OMR-NED (OMR Normalized Edit Distance).

Repository versions

The release state of each component repository, pinned for future reference. Each commit is the head of that repository's default branch at the time of release.

Repository	Branch	Commit
ossq-omr	omr-dev	7a17e45
omr-data-preprocessor	main	bdea0d1
sqomr	main	4926e69
lmxe	main	7eed5c9

Citation

If you use this benchmark, please cite:

@inproceedings{Kim2026sqomrbench,
    title     = {A Dataset and Benchmark for Optical Music Recognition of String Quartet Scores},
    author    = {Dongmin Kim and Brian Liu and Jose J. Valero-Mas and Dasaem Jeong},
    year      = 2026,
    booktitle = {Proceedings of the 27th International Society for Music Information Retrieval Conference (ISMIR)},
}

Depending on which components you use, please also cite:

If you use LMXE (the LMX linearizer/format, see modules/lmxe), the PyTorch implementation of the Zeus model, or the Zeus baseline numbers, please cite:

@inproceedings{mayer2024lmx,
    title     = {Practical End-to-End Optical Music Recognition for Pianoform Music},
    author    = {Mayer, Ji{\v{r}}{\'\i} and Straka, Milan and Haji{\v{c}} jr., Jan and Pecina, Pavel},
    year      = 2024,
    booktitle = {18th International Conference on Document Analysis and Recognition (ICDAR 2024)},
    pages     = {55--73},
    doi       = {10.1007/978-3-031-70552-6_4},
}

If you use the OSSQ-OMR dataset, please also cite the OpenScore String Quartets corpus it is built from:

@inproceedings{gotham_openscore_2023,
    title     = {The “{OpenScore} {String} {Quartet}” {Corpus}},
    author    = {Mark R. H. Gotham and Maureen Redbond and Bruno Bower and Peter Jonas},
    year      = 2023,
    booktitle = {Proceedings of the 10th {International} {Conference} on {Digital} {Libraries} for {Musicology}},
    pages     = {49--57},
    publisher = {ACM},
    address   = {Milan Italy},
    copyright = {All rights reserved},
    isbn      = {9798400708336},
    language  = {en},
    urldate   = {2023-11-16},
    month     = nov,
    doi       = {10.1145/3625135.3625155},
    url       = {https://dl.acm.org/doi/10.1145/3625135.3625155},
}

If you use the SMT baseline numbers, please cite:

@inproceedings{riosvila2024smt,
    title     = {Sheet Music Transformer: End-To-End Optical Music Recognition Beyond Monophonic Transcription},
    author    = {Ríos-Vila, Antonio and Calvo-Zaragoza, Jorge and Paquet, Thierry},
    year      = 2024,
    booktitle = {18th International Conference on Document Analysis and Recognition (ICDAR 2024)},
    pages     = {20--37},
    doi       = {10.1007/978-3-031-70552-6_2},
}

If you use the LEGATO baseline evaluated in this benchmark (submodule in sqomr, see Benchmark at a glance) or its baseline numbers, please also cite:

@inproceedings{yang2026legato,
    title     = {LEGATO: Large-scale End-to-end Generalizable Approach to Typeset OMR},
    author    = {Guang Yang and Victoria Ebert and Nazif Tamer and Brian Siyuan Zheng and Luiza Pozzobon and Noah A. Smith},
    year      = 2026,
    booktitle = {International Conference on Learning Representations (ICLR)},
    url       = {https://arxiv.org/abs/2506.19065},
}

License summary

Repo / component	License
ossq-omr (data)	CC0 1.0
omr-data-preprocessor	MIT
sqomr	MIT
lmxe	MIT
LEGATO baseline (submodule in sqomr)	MIT
— abc2xml.py within LEGATO	LGPL-3.0