Pith. sign in

REVIEW 4 major objections 5 minor 53 references

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

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper introduces OSSQ-OMR, the first dataset for multi-part optical music recognition, pairing 24,544 system and 98,172 staff images from 116 string quartet scores with token-aligned transcriptions, and records baselines of 3.6%…

desk verdict A genuinely first multi-part OMR dataset with a usable benchmark, but the manual alignment is the load-bearing assumption and the paper needs independent validation or a clear caveat. read the letter →

arxiv 2608.10978 v1 pith:UQKTTTBW submitted 2026-08-11 cs.CV cs.SD

Dongmin Kim , Brian Liu , Jose J. Valero-Mas , Dasaem Jeong This is my paper · ORCID
classification cs.CVcs.SD
keywords opticalmusicrecognitionmulti-partscoresstringquartetdatasetbenchmarkOMR-NEDLMXEvisualalignment
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Optical music recognition (OMR) has worked well for single-line and piano scores, but multi-part scores—where several instruments play simultaneously—have been held back by the absence of a dataset to train and test on. This paper introduces OSSQ-OMR, a dataset of 24,544 system-level and 98,172 staff-level images from 116 string quartet scores, each paired with a scanned edition and with token-aligned transcriptions in three symbolic encodings. The authors argue that because the dataset is public and comes with a benchmark protocol, multi-part OMR becomes a measurable and tractable problem. Their baseline experiments show the task is feasible, with best errors of 3.6% on synthetic and 5.9% on scanned images, while revealing that encoding and segmentation choices affect accuracy more than the choice of architecture.

What carries the argument

The load-bearing mechanism is the visual-alignment pipeline: the authors re-edit each OpenScore MuseScore file so that its MuseScore rendering aligns with the corresponding IMSLP scan at the token level, making the rendering and the symbolic transcription two views of the same ground truth. Around this core sit the YOLOv8-based segmentation pipeline that produces system and staff images, and the conversion pipeline that derives LMXE, **kern, and ABC from each corrected file with round-trip verification that removes samples failing structural or major musical checks. LMXE, an extension of Linearized MusicXML, is introduced to give a sequence-friendly, part-nesting encoding for multi-part scores.

What would settle it

Take a random sample of, say, 20 of the 93 aligned full-score scans, have an independent annotator transcribe each scan from scratch into a symbolic format, and compute token-level agreement (e.g., OMR-NED) between the independent transcription and the released ground truth; if the agreement is far above the claimed baseline errors, the ground truth—and all benchmark numbers—are compromised.

Watch

Extended reading notes

Core claim

The paper's central claim is that OSSQ-OMR is the first dataset dedicated to multi-part OMR and that it makes multi-part score transcription a feasible, benchmarkable task. The dataset pairs 93 scanned full scores from IMSLP with corrected MuseScore encodings that were manually re-edited so that the rendered image matches the scan at the token level; from these, synthetic and scanned images are segmented into systems and staves, and transcriptions are released in LMXE, **kern, and ABC. Baseline experiments with two representative models and nine tokenization schemes report OMR-NED as low as 3.6% on synthetic and 5.9% on scanned inputs, and show that part-level inputs outperform system-level inputs in every configuration, that LMXE beats the other encodings, and that the LSTM-based model degrades on scanned inputs roughly 2.6 times less than the Transformer-based model.

Load-bearing premise

The dataset's ground truth rides on the authors' manual visual alignment being correct at the token level: the 100+ hours of edits to 93 scores were not checked by an independent annotator or automated audit, so an alignment error would propagate into every paired image, transcription, and benchmark number.

Editorial extensions

If this is right

  • With a shared public benchmark, multi-part OMR results across models and encodings become directly comparable, so progress on this task can be measured rather than asserted.
  • The reported results imply that representation choice—encoding and segmentation level—matters more than the encoder–decoder architecture: practitioners should prioritize LMXE and part-level inputs over model swaps.
  • The scan-vs-synthetic gap (3.6% to 5.9% for the best configuration) quantifies the remaining domain-shift problem, giving future work a concrete target.
  • Because the corrected MuseScore sources are released as an auditable Git history, the dataset supports not only training but also studies of how notation and layout edits affect OMR output.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the alignment is as reliable as claimed, the same re-edit-and-render workflow could extend the corpus to other OpenScore genre sets, turning the entire OpenScore library into a multi-part OMR resource.
  • The consistent part-level advantage suggests current sequence models are not exploiting cross-part context; a model that jointly decodes the four staves of a system, rather than four independent sequences, is a natural next test.
  • The monotonic degradation of ABC-BPE with vocabulary size hints that for structured symbolic outputs, character-level or token-level supervision beats subword compression; this is a testable hypothesis that could generalize to other symbolic-sequence tasks.
  • A blind re-annotation of a random sample of the 93 aligned scores, performed without access to the authors' edits, would independently quantify the alignment's accuracy and is the direct way to check the dataset's ground truth.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces OSSQ-OMR, a dataset and benchmark for multi-part optical music recognition (OMR). The dataset provides 24,544 system images and 98,172 staff images drawn from 116 synthetic and 93 scanned string quartet scores, paired with symbolic transcriptions in three formats (LMXE, **kern, ABC). The images and transcriptions are derived from OpenScore String Quartet encodings that have been manually edited so that MuseScore renderings visually match the IMSLP scans. The benchmark evaluates two baseline architectures (Zeus and Sheet Music Transformer) across nine tokenization schemes and four score-level splits, reporting OMR-NED as low as 3.6% on synthetic and 5.9% on scanned inputs. The paper also includes an external baseline comparison with Legato.

Significance. If the manual alignment is trustworthy, OSSQ-OMR fills a clear gap in the OMR literature: it is the first dataset dedicated to multi-part score transcription, with a substantial number of images and multiple symbolic encodings. The release of the corrected MuseScore sources as auditable Git history is a concrete strength, and the external Legato comparison provides an independent anchor for evaluating the baselines. The benchmark protocol is clearly structured and the result that encoding choices substantially affect accuracy is useful for the community. However, the central validity of the paired image-transcription ground truth rests on an unvalidated manual alignment process, and the benchmark pool excludes a large, unquantified fraction of the released data due to height and length caps. These issues must be addressed before the dataset and benchmark can be fully relied upon.

major comments (4)
  1. [Section 2.2, Table 1] The claim that the rendered MuseScore images match the IMSLP scans 'at the token level' is not supported by any quantitative validation. The paper reports 100+ hours of iterative editing but no inter-annotator agreement, no automated render-to-scan correspondence check, and no independent audit. Since Table 1 records 7,183 pitch edits and 1,948 slur edits across a subset of the files, the prior for human error or overcorrection is nontrivial. Because the corrected MuseScore files are the single source for both the synthetic training images and the symbolic ground truth for scanned images, any alignment error propagates into the training distribution and the evaluation reference. The released Git history makes the edits auditable but does not by itself establish correctness. Please add a validation study, for example a second annotator on a random sample or an automated image-to-symbol correspondence metric, and provide an operational definition of 'token-level' alignment.
  2. [Section 3.3, Table 3] The benchmark pool is the intersection of samples passing the height cap and all per-encoding length caps, but the paper does not report how many samples are excluded by each filter or analyze the coverage bias. Table 3 is internally inconsistent: for system-level, train 13,552±128 and valid 1,611±1 already exceed the reported 'Total (synth)' of 10,016, so the table's meaning is unclear. The release counts in Table 2 (e.g., 13,240 synthetic system images) are far larger than the benchmark pool appears to be, which may imply that a large fraction of long systems or verbose encodings is absent from the evaluation. Please report the exact filtering funnel (raw -> height cap -> each encoding cap -> intersection) and characterize the excluded samples, since the baselines only speak to the filtered subset.
  3. [Section 4.1, Section 4.2, Table 4] The headline robustness result ('the LSTM-based baseline degrading on scanned inputs roughly 2.6 times less than the Transformer-based baseline') is confounded: Zeus and SMT differ simultaneously in architecture, learning rate (0.001 with cosine schedule vs 0.0001 without), and, for Zeus, modified hidden sizes. The paper acknowledges that the architectural source of the gap is not isolated, but the abstract and conclusions state the result as an architecture comparison. Please either provide a controlled comparison with matched training hyperparameters or soften the attribution to the models as configured.
  4. [Abstract and Section 2.2, Section 2.5] The abstract states that 'all images visually aligned to their transcriptions,' but Section 2.2 says the alignment process 'was exclusively applied to the 93 full-score scans.' If scanned images from the 17 part-books and 6 manuscripts are included in the release without the same alignment, the claim is overbroad; if they are not included, the dataset's scanned coverage is only 93 of the 116 scores. Please clarify the status of part-book and manuscript scans and align the abstract with the actual alignment scope.
minor comments (5)
  1. [Table 3] The meaning of the 'Total (synth)' and 'Total (scan)' rows is unclear; please rewrite the table to distinguish per-split counts from overall pool sizes.
  2. [Section 4.2] The robustness comparison reports both relative degradation (39% vs 100% increase) and absolute increases (3.6 pp vs 6.7 pp); the '2.6×' factor should be defined exactly to avoid ambiguity.
  3. [Section 3.4] The time-signature injection for **kern and ABC makes time-signature accuracy incomparable across segmentation levels; this is stated but should be reflected in the conclusions when comparing encodings.
  4. [References] Reference [28] points to MuseScore.com rather than the MuseScore software; please correct the citation.
  5. [Section 4.3] The comparison of Legato's system-wise OMR-NED on OSSQ-OMR to its previously reported page-wise result is only indicative because the segmentation levels differ; please state this explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the benchmark evaluation is self-contained and anchored by an external baseline; no predicted quantity reduces by construction to a fitted input or self-citation.

full rationale

This paper introduces a dataset and benchmark for multi-part OMR; it contains no derivation chain in which a predicted quantity is defined in terms of the data it is supposed to predict. The ground-truth transcriptions are produced by a documented manual visual-alignment process, and while its correctness is not independently validated, that is a data-quality limitation rather than circularity: the alignment is not derived from the OMR models, nor are the OMR-NED numbers fitted to it. The staff-height regression model cited from prior work [29] is used as-is as a preprocessing component; even though the citation involves overlapping authors, it is not load-bearing for the headline feasibility claim, which rests on held-out test splits of newly paired scanned and synthetic data. The authors also design LMXE and then show it performs well, which is a potential design-validation bias but not circularity, because performance is measured against an independent symbolic comparison via musicdiff and because the external Legato baseline provides an independent anchor. No equation in the paper reduces to its own inputs, and no self-citation is invoked to forbid alternatives or to force the reported result. Accordingly, the circularity score is 0.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The central claims depend on manually chosen filtering caps, a new encoding format, and assumptions about evaluation and alignment validity. The most load-bearing item is the unverified manual alignment, followed by the sample exclusion policy that shapes all benchmark numbers.

free parameters (4)
  • Per-encoding sequence length caps = LMXE 800/430, LMXE-P 790, EKERN 1100/390, BEKERN 1400/540, CABC 1170/500, ABC-BPE 340/160, ABC-BPE-1024 430/210…
    Chosen by hand to balance compute and coverage; directly determines which samples enter the benchmark pool (Section 3.2, 3.3).
  • System image height cap = 256 pixels
    Excludes taller system images to fit baseline model input constraints (Section 3.3); affects the benchmark sample pool.
  • ABC-BPE vocabulary sizes = 4096, 1024, 512, 256
    Ablation levels chosen to test BPE granularity; results show monotonic degradation, but vocabulary size is a design choice (Section 3.2).
  • Baseline hyperparameters = Zeus lr=0.001, SMT lr=0.0001, 100k updates, batch 64/256
    Learning rates and schedules differ between Zeus and SMT, so the architecture comparison is confounded (Section 4.1).
assumptions (4)
  • domain assumption OMR-NED faithfully measures transcription quality
    Used as the only evaluation metric (Eq. 1, Section 3.4); if OMR-NED misses musically relevant errors, benchmark conclusions may not transfer.
  • domain assumption Round-trip MusicXML conversion detects symbol-level errors adequately
    Used to exclude conversion-failure samples (Section 2.4); assumes the OMR-NED categories cover all important error types.
  • domain assumption Manual visual alignment by the authors is correct
    Ground truth quality rests on 100+ hours of manual editing without inter-annotator agreement or independent validation (Section 2.2).
  • domain assumption MuseScore 3.6.2 default rendering is a representative synthetic proxy for scans
    Synthetic images are rendered with default engraving and treated as a distinct test set; differences from scans are attributed to domain shift (Section 2.3).
invented entities (1)
  • LMXE (Extended Linearized MusicXML)
    purpose: Compact, sequence-friendly multi-part encoding used as the paper's primary tokenization
    Introduced in Section 2.4; no external implementation or evaluation outside this paper, so its advantages are self-assessed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Dataset and Benchmark for Optical Music Recognition of String Quartet Scores." pith.science (2026). https://pith.science/paper/UQKTTTBW

@misc{pith2026260810978,
  author       = {Pith},
  title        = {Pith review of: A Dataset and Benchmark for Optical Music Recognition of String Quartet Scores},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UQKTTTBW}},
  note         = {Machine review of arXiv:2608.10978}
}
read the original abstract

Optical music recognition (OMR) transcribes music scores into digital formats. While the field has advanced significantly on monophonic and piano-form scores, multi-part score transcription remains underexplored, largely due to the absence of a suitable dataset. We introduce OpenScore String Quartet for Optical Music Recognition (OSSQ-OMR), the first dataset dedicated to multi-part OMR. Built on the OpenScore String Quartet corpus, OSSQ-OMR pairs digitally encoded scores with their original scanned editions from IMSLP, with all images visually aligned to their transcriptions. The dataset is released with score images at system and staff levels, and paired transcriptions in three encoding formats: Extended Linearized MusicXML (LMXE), **kern, and ABC. In total, OSSQ-OMR contains 24,544 system images and 98,172 staff images drawn from 116 string quartet scores. We accompany the dataset with a benchmark protocol and baseline results from two representative OMR models, evaluated across four random score-level splits with mutually exclusive test sets. Baselines reach OMR-NED as low as 3.6% on synthetic and 5.9% on scanned inputs; results reveal substantial effects of encoding and segmentation choices, with the LSTM-based baseline degrading on scanned inputs roughly 2.6 times less than the Transformer-based baseline.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 50 canonical work pages

  1. [1]

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

    INTRODUCTION Optical music recognition (OMR) refers to the field that studies computational methods for automatically transcrib- ing music score images into structured digital formats [1]. It plays a key role in the preservation of musical cul- tural heritage while enabling its exploitation and analy- sis through Music Information Retrieval (MIR) techniqu...

    work page Pith review arXiv 2026

  2. [2]

    To extend it, we retrieve the corresponding scanned scores from IMSLP [26] and categorize them as follows: •Full score(93 scores, 80%): all four parts are shown on a single system

    THE OSSQ-OMR DATASET 2.1 Source corpus and scan collection We build the dataset from the OpenScore String Quartet corpus [25], comprising 116 string quartet scores encoded as 122 MuseScore files (with two multi-movement works split by movement), along with rendered image versions. To extend it, we retrieve the corresponding scanned scores from IMSLP [26] ...

    work page 1994

  3. [3]

    MuseScore files are exported to MusicXML using MuseScore [28]. The resulting MusicXML under- goes a cleansing stage to remove invisible elements and their associated backup/forward directives, which would otherwise produce visually absent but symboli- cally present content

    work page

  4. [4]

    The cleaned MusicXML is converted to system-wise LMXE using a custom tool based on [30]

    work page

  5. [5]

    From system-wise LMXE, we derive system-wise **kernand system-wise ABC by converting back to MusicXML and then forward usingconverter21. 2

    work page

  6. [6]

    Part-wise encodings are derived as follows: part-wise LMXE is produced directly from system-wise LMXE using the custom tool; part-wise **kernis produced from system-wise **kernusingkernpy; part-wise ABC is produced by converting part-wise LMXE to MusicXML and then to ABC viaconverter21. Verification and exclusions.Each conversion step ex- cept the initial...

    work page

  7. [7]

    The remainder of the section describes the experimental details: baseline models, tok- enization schemes, data splits, and evaluation metrics

    BENCHMARK PROTOCOL We accompany OSSQ-OMR with benchmark experiments to analyze the capabilities of current state-of-the-art OMR models on multi-part scores. The remainder of the section describes the experimental details: baseline models, tok- enization schemes, data splits, and evaluation metrics. 2 https://github.com/gregchapman-dev/ converter21 Source ...

    work page

  8. [8]

    Batch sizes are 64 at the system level and 256 at the part level (reflecting the∼4×larger sample count)

    BASELINE EXPERIMENTS 4.1 Training setup All models are trained with the AdamW [42] optimizer for 100,000 updates, with auto-regressive validation every 5,000 steps monitoring OMR-NED on the validation set. Batch sizes are 64 at the system level and 256 at the part level (reflecting the∼4×larger sample count). Learning rates are set to 0.001 with a cosine ...

    work page

Show all 53 references
  1. [9]

    CONCLUSIONS AND FUTURE WORK We introduced OSSQ-OMR, the first dataset dedicated to multi-part OMR, together with a benchmark protocol and baseline results from two representative architectures. The benchmark shows that representation choices matter more than architecture: LMXE...

  2. [10]

    Advanced GPU Utilization Support Program

    ACKNOWLEDGMENTS This work was supported by the Ministry of Education of the Republic of Korea and the National Research Founda- tion of Korea (NRF-2024S1A5C3A03046168), and by the “Advanced GPU Utilization Support Program” funded by the Government of the Republic of Korea (Min...

  3. [11]

    All dataset construction, experimental design, training, evalu- ation, and analysis were carried out by the authors, who verified all reported results

    AI USAGE STATEMENT AI tools were used only for minor language editing, such as paraphrasing for readability and grammar checking. All dataset construction, experimental design, training, evalu- ation, and analysis were carried out by the authors, who verified all reported results

  4. [12]

    Under- standing optical music recognition,

    J. Calvo-Zaragoza, J. H. Jr, and A. Pacha, “Under- standing optical music recognition,”ACM Computing Surveys (CSUR), vol. 53, no. 4, pp. 1–35, 2020

    2020

  5. [13]

    Optical mu- sic imaging: music document digitisation, recognition, evaluation, and restoration,

    G. Jones, B. Ong, I. Bruno, and K. Ng, “Optical mu- sic imaging: music document digitisation, recognition, evaluation, and restoration,” inInteractive multimedia music technologies. IGI Global Scientific Publishing, 2008, pp. 50–79

    2008

  6. [14]

    E. R. Miranda,Handbook of artificial intelligence for music. Springer, 2021

    2021

  7. [15]

    Calvo-Zaragoza, J

    J. Calvo-Zaragoza, J. Martinez-Sevilla, C. Peñarrubia, and A. Ríos Vila,Optical Music Recognition: Recent Advances, Current Challenges, and Future Directions, 08 2023, pp. 94–104

    2023

  8. [16]

    Deep learning for optical music recognition: A review,

    F. J. Castellanos, A. J. Gallego, and I. Fujinaga, “Deep learning for optical music recognition: A review,” TechRxiv, vol. 2025, no. 0228, 2025

    2025

  9. [17]

    A neural approach for full-page optical music recog- nition of mensural documents,

    F. J. Castellanos, J. Calvo-Zaragoza, and J. M. Iñesta, “A neural approach for full-page optical music recog- nition of mensural documents,” inProceedings of the 21th International Society for Music Information Re- trieval Conference (ISMIR), 2020, pp. 558–565

    2020

  10. [18]

    Region-based layout analysis of music score images,

    F. J. Castellanos, C. Garrido-Munoz, A. Ríos-Vila, and J. Calvo-Zaragoza, “Region-based layout analysis of music score images,”Expert Systems with Applica- tions, vol. 209, p. 118211, 2022

    2022

  11. [19]

    Optical medieval music recognition—a complete pipeline for historic chants,

    A. Hartelt, T. Eipert, and F. Puppe, “Optical medieval music recognition—a complete pipeline for historic chants,”Applied Sciences, vol. 14, no. 16, 2024

    2024

  12. [20]

    A region-based approach for layout analysis of music score images in scarce data scenarios,

    F. J. Castellanos, J. P. Martinez-Esteso, A. Galán- Cuenca, and A. J. Gallego, “A region-based approach for layout analysis of music score images in scarce data scenarios,” inInternational Confernece on Document Analysis and Recognition, Cham, 2024, pp. 58–75

    2024

  13. [22]

    Sheet music transformer: End-to-end optical music recog- nition beyond monophonic transcription,

    A. Ríos-Vila, J. Calvo-Zaragoza, and T. Paquet, “Sheet music transformer: End-to-end optical music recog- nition beyond monophonic transcription,” inInterna- tional Conference on Document Analysis and Recogni- tion. Springer, 2024, pp. 20–37

    2024

  14. [23]

    Sheet music transformer++: End-to-end full- page optical music recognition for pianoform sheet music,

    A. Ríos-Vila, J. Calvo-Zaragoza, D. Rizo, and T. Pa- quet, “Sheet music transformer++: End-to-end full- page optical music recognition for pianoform sheet music,”arXiv preprint arXiv:2405.12105, 2024

    2024 arXiv

  15. [24]

    An implicit layout-aware transformer for full- page end-to-end optical music recognition,

    A. Rios-Vila, E. Fuentes-Martinez, and F. J. Castel- lanos, “An implicit layout-aware transformer for full- page end-to-end optical music recognition,”Interna- tional Journal of Multimedia Information Retrieval, vol. 14, no. 4, p. 34, 2025

    2025

  16. [25]

    End-to-end full-page optical music recognition for pianoform sheet music,

    A. Ríos-Vila, J. Calvo-Zaragoza, D. Rizo, and T. Pa- quet, “End-to-end full-page optical music recognition for pianoform sheet music,”International Journal of Computer Vision, vol. 134, no. 2, p. 49, 2026

    2026

  17. [26]

    Optical music recognition by long short-term mem- ory networks,

    A. Baró, P. Riba, J. Calvo-Zaragoza, and A. Fornés, “Optical music recognition by long short-term mem- ory networks,” inInternational Workshop on Graphics Recognition. Springer, 2017, pp. 81–95

    2017

  18. [27]

    End-to-end optical music recognition using neural networks,

    J. Calvo-Zaragoza, J. J. Valero-Mas, and A. Pertusa, “End-to-end optical music recognition using neural networks,” inProceedings of the 18th International So- ciety for Music Information Retrieval Conference (IS- MIR), 2017, pp. 23–27

    2017

  19. [28]

    A sequence-to-sequence frame- work based on transformer with masked language model for optical music recognition,

    C. Wen and L. Zhu, “A sequence-to-sequence frame- work based on transformer with masked language model for optical music recognition,”IEEE Access, vol. 10, pp. 118 243–118 252, 2022

    2022

  20. [29]

    An empirical evaluation of end-to-end polyphonic optical music recognition,

    S. Edirisooriya, H. Dong, J. J. McAuley, and T. Berg- Kirkpatrick, “An empirical evaluation of end-to-end polyphonic optical music recognition,” inProceedings of the 22nd International Society for Music Informa- tion Retrieval Conference (ISMIR), 2021, pp. 167–173

    2021

  21. [30]

    Camera-primus: Neu- ral end-to-end optical music recognition on realistic monophonic scores

    J. Calvo-Zaragoza and D. Rizo, “Camera-primus: Neu- ral end-to-end optical music recognition on realistic monophonic scores.” inProceedings of the 19th Inter- national Society for Music Information Retrieval Con- ference (ISMIR), 2018, pp. 248–255

    2018

  22. [31]

    Two (note) heads are better than one: Pen-based multimodal interaction with music scores

    J. Calvo-Zaragoza, D. Rizo, and J. M. I. Quereda, “Two (note) heads are better than one: Pen-based multimodal interaction with music scores.” inProceedings of the 17th International Society for Music Information Re- trieval Conference (ISMIR), 2016, pp. 509–514

    2016

  23. [32]

    A diplomatic edition of il lauro secco: Ground truth for OMR of white mensural notation,

    E. Parada-Cabaleiro, A. Batliner, and B. W. Schuller, “A diplomatic edition of il lauro secco: Ground truth for OMR of white mensural notation,” inProceedings of the 20th International Society for Music Information Retrieval Conference (ISMIR), 2019, pp. 557–564

    2019

  24. [33]

    Encoded spanish music heritage through verovio: The online platforms fondo de música tradicional imf–csic and books of hispanic polyphony,

    E. Ros-Fábregas, S. Münnich, and D. Rizo, “Encoded spanish music heritage through verovio: The online platforms fondo de música tradicional imf–csic and books of hispanic polyphony,” inProceedings of the Music Encoding Conference, 2021

    2021

  25. [34]

    Digi- tization of choirbooks in guatemala,

    M. E. Thomae, J. E. Cumming, and I. Fujinaga, “Digi- tization of choirbooks in guatemala,” inProceedings of the 9th International Conference on Digital Libraries for Musicology (DLfM), New York, NY , USA, 2022, p. 19–26

    2022

  26. [35]

    End-to-end optical music recognition for pianoform sheet music,

    A. Ríos-Vila, D. Rizo, J. M. Iñesta, and J. Calvo- Zaragoza, “End-to-end optical music recognition for pianoform sheet music,”International Journal on Doc- ument Analysis and Recognition (IJDAR), vol. 26, no. 3, pp. 347–362, 2023

    2023

  27. [36]

    The “OpenScore String Quartet

    M. Gotham, M. Redbond, B. Bower, and P. Jonas, “The “OpenScore String Quartet” Corpus,” inProceedings of the 10th International Conference on Digital Li- braries for Musicology (DLfM). ACM, 2023, pp. 49– 57

    2023

  28. [37]

    (2026) International music score library project (petrucci music library)

    IMSLP. (2026) International music score library project (petrucci music library). [Online]. Available: https://imslp.org/

    2026

  29. [38]

    Ultralytics yolov8,

    G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics yolov8,” pp. 1–6, 2023. [Online]. Available: https: //github.com/ultralytics/ultralytics

    2023

  30. [39]

    MuseScore.com | The world’s largest free sheet music catalog and community — mus- escore.com,

    Musescore, “MuseScore.com | The world’s largest free sheet music catalog and community — mus- escore.com,” https://musescore.com/, [Accessed 04- 14-2025]

    2025

  31. [40]

    U-MusT: A unified frame- work for cross-modal translation of score images, sym- bolic music, and performance audio,

    J. Jung, D. Kim, S. Lee, S. Cho, H. Soh, I. Bukey, C. Donahue, and D. Jeong, “U-MusT: A unified frame- work for cross-modal translation of score images, sym- bolic music, and performance audio,”IEEE Trans- actions on Audio, Speech and Language Processing, vol. 34, pp. 1876–1891, 2026

    2026

  32. [41]

    Practi- cal end-to-end optical music recognition for pianoform music,

    J. Mayer, M. Straka, J. Haji ˇc, and P. Pecina, “Practi- cal end-to-end optical music recognition for pianoform music,” inDocument Analysis and Recognition - IC- DAR 2024. Cham: Springer Nature Switzerland, 2024, pp. 55–73

    2024

  33. [42]

    Huron,Humdrum and Kern: selective feature en- coding

    D. Huron,Humdrum and Kern: selective feature en- coding. Cambridge, MA, USA: MIT Press, 1997, p. 375–401

    1997

  34. [43]

    The abc music standard 2.1 (dec 2011),

    C. Walshaw, “The abc music standard 2.1 (dec 2011),” 2011

    2011

  35. [44]

    LEGATO: Large-scale end-to-end generalizable approach to typeset OMR,

    G. Yang, V . Ebert, N. Tamer, B. S. Zheng, L. A. Pozzobon, and N. A. Smith, “LEGATO: Large-scale end-to-end generalizable approach to typeset OMR,” inProceedings of The 14th International Conference on Learning Representations (ICLR), 2026

    2026

  36. [45]

    Sheet Music Benchmark: Stan- dardized optical music recognition evaluation,

    J. C. Martinez-Sevilla, J. Cerveto-Serrano, N. N. Luna-Barahona, G. Chapman, C. Sapp, D. Rizo, and J. Calvo-Zaragoza, “Sheet Music Benchmark: Stan- dardized optical music recognition evaluation,” inPro- ceedings of the 26th International Society for Music Information Retrieval...

    2025

  37. [46]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inIEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE Computer Society, 2016, pp. 770–778

    2016

  38. [47]

    A convnet for the 2020s,

    Z. Liu, H. Mao, C. Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A convnet for the 2020s,” inIEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR). IEEE, 2022, pp. 11 966–11 976

    2022

  39. [48]

    kernpy: a humdrum **kern oriented python package for optical music recognition tasks,

    J. Cerveto-Serrano, D. Rizo, and J. Calvo-Zaragoza, “kernpy: a humdrum **kern oriented python package for optical music recognition tasks,” inProceedings of the Music Encoding Conference, 2025

    2025

  40. [49]

    Neural ma- chine translation of rare words with subword units,

    R. Sennrich, B. Haddow, and A. Birch, “Neural ma- chine translation of rare words with subword units,” in Proceedings of the 54th Annual Meeting of the Associ- ation for Computational Linguistics (ACL), 2016

    2016

  41. [50]

    music21: A toolkit for computer-aided musicology and symbolic music data,

    M. S. Cuthbert and C. Ariza, “music21: A toolkit for computer-aided musicology and symbolic music data,” inProceedings of the 11th International Society for Music Information Retrieval Conference (ISMIR), 2010, pp. 637–642

    2010

  42. [51]

    A diff procedure for music score files,

    F. Foscarin, F. Jacquemard, and R. Fournier-S’niehotta, “A diff procedure for music score files,” inProceed- ings of the 6th International Conference on Digital Li- braries for Musicology (DLfM), 2019, pp. 58–64

    2019

  43. [52]

    Transcoda: End-to-end zero-shot optical music recognition via data-centric synthetic training,

    D. Dratschuk and P. Swoboda, “Transcoda: End-to-end zero-shot optical music recognition via data-centric synthetic training,”arXiv preprint arXiv:2605.10835, 2026

    2026 arXiv

  44. [53]

    Decoupled weight de- cay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight de- cay regularization,” in7th International Conference on Learning Representations (ICLR), 2019

    2019

  45. [54]

    PDMX: A large-scale public domain mu- sicxml dataset for symbolic music processing,

    P. Long, Z. Novack, T. Berg-Kirkpatrick, and J. J. McAuley, “PDMX: A large-scale public domain mu- sicxml dataset for symbolic music processing,” in IEEE International Conference on Acoustics, Speech and Signal Processing, (ICASSP), 2025, pp. 1–5

    2025

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.