Sheet Music Transformer++: End-to-End Full-Page Optical Music Recognition for Pianoform Sheet Music

Full-text preprint.

About • How To Use • Citations • License

About

This GitHub repository contains the implementation of the Sheet Music Transformer ++ (SMT), the upgraded version of the Sheet Music Transformer model for full-page pianoform music sheet transcription. Unlike traditional approaches that primarily resort this challenge by implementing layout analysis techniques with end-to-end transcription, the SMT ++ model offers a image-to-sequence solution for transcribing these scores directly from images. To do so, this model is trained through a progressive curriculum learning strategy with synthetic generation.

⚠️ Please bear in mind that, although some results have been published, this is still an work-in-progress project, bugs may be found

How to use

Usage instructions and data publication coming (hopefully) soon!

Citations

@misc{RiosVila:2024:SMTplusplus,
      title={Sheet Music Transformer ++: End-to-End Full-Page Optical Music Recognition for Pianoform Sheet Music}, 
      author={Antonio Ríos-Vila and Jorge Calvo-Zaragoza and David Rizo and Thierry Paquet},
      year={2024},
      eprint={2405.12105},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}