AEmotionStudio/smt-omr-models

SMT-fp — full-page Sheet Music Transformer (MAESTRO mirror)

Mirror of PRAIG's full-page Optical Music Recognition checkpoints, laid out one subdirectory per variant for MAESTRO's download-on-demand loader:

subdir	upstream	material
grandstaff/	PRAIG/smt-fp-grandstaff	engraved piano (two-staff) scores
polish-scores/	PRAIG/smt-fp-polish-scores	historical printed scores (Polish digital libraries)
mozarteum/	PRAIG/smt-fp-mozarteum	Mozart-era engravings (Digital Mozart Edition)

Each variant is config.json (vocabulary embedded) + model.safetensors, byte-identical to upstream (sha256-verified at mirror time). Model class: SMTModelForCausalLM — ConvNeXt encoder + autoregressive kern decoder from antoniorv6/SMT (MIT).

License

MIT, as published by PRAIG (Pattern Recognition and Artificial Intelligence Group, University of Alicante). This mirror adds no restrictions.

Citation

@article{RiosVila2024SMT,
  title   = {Sheet Music Transformer: End-To-End Optical Music Recognition Beyond Monophonic Transcription},
  author  = {R{\'i}os-Vila, Antonio and Calvo-Zaragoza, Jorge and Paquet, Thierry},
  journal = {International Conference on Document Analysis and Recognition (ICDAR)},
  year    = {2024}
}

Mirrored for MAESTRO by AEmotionStudio.