Research map / Architectures and efficiency / RNNs and normalization
LEGATO: Large-scale End-to-end Generalizable Approach to Typeset OMR
Guang Yang, Victoria Ebert, Nazif Can Tamer, Brian Siyuan Zheng, Luiza Amador Pozzobon, Noah A. Smith
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Legato (frozen pretrained vision encoder + ABC decoder trained on PDMX-Synth) instead of SMT++ | full-page OMR · Camera OpenScore String Quartets | TEDn: 98.6 -> 62.2 | Improved |
| use Legato instead of SMT++ on real IMSLP piano scans | full-page OMR · IMSLP Piano Scores | TEDn: 97.7 -> 29.5 | Improved |
| switch SMT++ output from **kern to ABC with BPE tokenizer (ablation) | SMT++ on GrandStaff · IMSLP Piano Scores | TEDn: 97.7 -> 74.6 | Improved |
| enlarge SMT++ to 177M parameters (ablation) | SMT++ on ABC GrandStaff · IMSLP Piano Scores | TEDn: 74.6 -> 73.0 | No clear effect |
| train enlarged SMT++ on larger PDMX-Synth data (ablation) | SMT++_large · IMSLP Piano Scores | TEDn: 73.0 -> 80.5 | Hurt |
| use frozen pretrained vision encoder with Legato architecture and PDMX-Synth instead of SMT++_large (ablation) | OMR model · IMSLP Piano Scores | TEDn: 80.5 -> 29.5 | Improved |
| use larger decoder (Legato versus Legato_small) (ablation) | Legato · IMSLP Piano Scores | TEDn: 75.7 -> 29.5 | Improved |
| add three in-context ABC examples to GPT-5 prompt (ablation) | GPT-5 · IMSLP Piano Scores | TEDn: 96.7 -> 74.7 | Improved |
| prompt GPT-5 for MusicXML instead of ABC (ablation) | GPT-5 · IMSLP Piano Scores | TEDn: 96.7 -> 86.6 | Improved |
| prompt Gemini 2.5 Pro with in-context ABC examples (ablation) | Gemini 2.5 Pro · IMSLP Piano Scores | TEDn: 87.8 -> 78.6 | Improved |
| evaluate on violin parts only with Verovio rendering, favoring SMT++ | Legato versus SMT++ · Rendered String Quartets Violin Parts | TEDn: 95.0 -> 7.7 | Improved |
| use Legato_small instead of SMT++ on violin parts | Legato_small · Rendered String Quartets Violin Parts | CER kern: 30.7 -> 50.6 | Hurt |
Closest papers by meaning
- Transcoda: End-to-End Zero-Shot Optical Music Recognition via Data-Centric Synthetic Training (NeurIPS 2026 · similarity 0.91)
- Toward Multimodal Sheet Music Recognition and Understanding (NeurIPS 2026 · similarity 0.91)
- Rubato: Transcribing Piano Music with Timestamps (NeurIPS 2026 · similarity 0.86)
- Aria-MIDI: A Dataset of Piano MIDI Files for Symbolic Music Modeling (ICLR 2025 · similarity 0.85)
- MuPT: A Generative Symbolic Music Pretrained Transformer (ICLR 2025 · similarity 0.83)
- UniMuMo: Unified Text, Music, and Motion Generation (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
