Sample-efficient end-to-end Optical Music Recognition through Active Learning
Organizations: University of Alicante, Carretera San Vicente del Raspeig s/n, San Vicente del Raspeig, 03690, Spain
Abstract
End-to-end approaches to Optical Music Recognition (OMR) have recently shown promising results, but their performance often degrades on unseen data. A common solution is to pretrain and fine-tune models with annotated samples from the target domain, improving adaptation to new distributions. However, this process typically requires extensive manual annotation to yield competitive models, making it costly and time-consuming. In this work, we explore Active Learning to reduce this burden by selecting informative samples for labeling in low-resource scenarios, where the amount of manual annotation that can be performed is limited. We evaluate both uncertainty- and diversity-based sampling strategies on staff-level and full-page corpora. Results show that confidence-based sampling performs comparably to oracle selection (based on true error), and that each scenario benefits from a tailored strategy.
Figures & tables
| Dataset | Symbols | Sequence length | # of samples | Handwritten or Printed | ||||||
| Unique | Total | Min | Average | Max | ||||||
| Staff-level | ||||||||||
| Capitan | 323 | 17112 | 8 | 24.3 | 35 | 704 | H | |||
| Guatemala | 252 | 78558 | 4 | 24.1 | 49 | 3263 | H | |||
| Magnificat | 181 | 40085 | 2 | 29.5 | 53 | 1361 | P | |||
| Mottecta | 199 | 43192 | 5 | 23.4 | 41 | 1845 | P | |||
| Dataset | Zero-shot - | Budget | - Oracle - | ||||
| 10 | 25 | 50 | 100 | ||||
| Staff-level | |||||||
| Capitan | 95.70 | 55.61 | 45.08 | 30.80 | 21.55 | 0 8.61 | |
| Guatemala | 84.74 | 36.67 | 24.13 | 19.23 | 14.69 | 0 2.44 | |
| Magnificat | 42.20 | 19.81 | 14.75 | 0 8.73 | 0 5.65 | 0 2.95 | |
| Mottecta | 38.32 | 24.89 | 17.20 | 12.73 | 0 9.14 | 0 5.92 | |
| Budget | Uncertainty | ||||
| Lowest | Near | Near Mean | Near | Highest | |
| 10 | - 1.72 | -4.20 | -0.94 | 1.93 | 0 8.68 |
| 25 | -1.46 | -0.39 | -1.60 | 7.01 | 0 7.22 |
| 50 | -2.75 | -2.02 | -2.22 | 3.85 | 10.49 |
| 100 | -1.59 | -1.82 | - 0.36 | 1.86 | 0 1.62 |
Optical Music Recognition
Optical Music Recognition converts images of printed or handwritten scores into structured musical notation for search, analysis, and playback.
See all 19 papers