MasonDill/Monophonic_OMR

tf-deep-omr

TensorFlow model to perform end-to-end Optical Music Recognition on monophonic scores through Convolutional Recurrent Neural Networks and CTC-based training.

Citation

End-to-End Neural Optical Music Recognition of Monophonic Scores

@Article{Calvo-Zaragoza2018,
  AUTHOR = {Calvo-Zaragoza, Jorge and Rizo, David},
  TITLE = {End-to-End Neural Optical Music Recognition of Monophonic Scores},
  JOURNAL = {Applied Sciences},
  VOLUME = {8},
  YEAR = {2018},
  NUMBER = {4},
  ARTICLE NUMBER = {606},
  URL = {http://www.mdpi.com/2076-3417/8/4/606},
  ISSN = {2076-3417},
  DOI = {10.3390/app8040606}
}

Corpora

This repository is intended for the Printed Images of Music Staves (PrIMuS) dataset.

PrIMuS can be donwloaded from https://grfia.dlsi.ua.es/primus/

Source Code

The source code used to generate the models, and examples on how to use it, can be found at https://github.com/MasonDill/tf-end-to-end

This code was altered from https://github.com/OMR-Research/tf-end-to-end

Other Models

Models from the origonal author can be found here:

• Agnostic model
• Semantic model

These models were the result of the traning process for one of the folds of the 10-fold cross-validation considered in the paper.

Let's see an example for the sample from PrIMuS provided in Data/Example:

This sample belongs to the test set of the aforementioned fold, so it was not seen by the networks during their training stage.