0.1

SMT for Full-page transcription V0.1

At last, I can publish the first finished version of the Sheet Music Transformer for full-page Optical Music recognition. Several changes have been introduced from the original repository (for the sake of usability, mostly). Among these ones, the most important ones are the following:

Research:

• New version of the preprint (sent already for review & publication).
• Two new benchmark datasets: Mozarteum and Digital Polish Scores.
• Performance metrics and comparison with a commercial OMR tool.

### Technical

• Release of the SMT weights through HuggingFace Transformers 🤗. You are now able to use the finetuned weights of the SMT through the Transformers library. Currently, only featured Mozarteum and Digital Polish Scores, Full-page GrandStaff wil come soon.
• Release of the datasets. Datasets are also vailable through the HuggingFace Datasets library. However only Full-page GrandStaff is publicly accessible. Mozarteum is kept private until publication and requests have to be done to get the Digital Polish Scores dataset. Hopefully, once the paper is published, we will be able to release both datasets for public access.

Of course, if any bug, issues, or doubts, please contact me for support.

Thanks for your interest!