OMR Datasets Versions Save

Added the download of the DeepScores V1 dataset (subset with extended vocabulary) to the available datasets. Also slightly rewrote the downloader, so you can specify a custom URL to any zip file that will be downloaded and extracted, in case your dataset is hosted elsewhere.

Fixed an error in the dependency management of setup.py which prevented normal installation. Also changing to semantic versioning with three numbers Major.Minor.Revision.

Updated MuscimaPlusPlusSymbolImageGenerator to work with MUSCIMA++ 2.0.

Added quality-of-life improvement suggested by @yvan674 to make importing common classes such as the downloader easier.

This release is a major new version that is incompatible with previous versions.

The project was dramatically simplified to ease common tasks such as downloading datasets two become much more user-friendly.

Removed mostly unused code and re-organized project structure and documentation.

Release of MaskImageGenerator to support three types of masks being generated from the MUSCIMA++ dataset, Version 2.0:

• Instance Segmentation of Staff Lines
• Instance Segmentation of Staff Blobs
• Semantic Segmentation of all Node objects

This release includes support for the new MUSCIMA++ v2.0 dataset.

In this release, we added a derived datasest from the MUSCIMA++ dataset, that contains annotations for staves, system measures and stave measures.

Further changes:

• Improved the documentation, so readthedocs actually renders the content properly.
• Significantly improved the download-progress reporting by resorting to TQDM instead of simply printing lines.
• Dropping Python 3.5 support, because we're using string-interpolation

Moving all datasets to Github and mirroring existing datasets to improve performance and ensure they will be preserved, even if the original sites shut down.

Mirroring:

• AudiverisOMR
• CVC-MUSCIMA - (Write identification, Staff removal, and Multi-Condition-Aligned).
• HOMUS 1.0 and 2.0
• MUSCIMA++ v1.0 including the images of the CVC-MUSCIMA dataset, that were annotated (CVC_MUSCIMA_PP_Annotated-Images.zip, Measure-Annotations)
• OpenOMR
• PrintedMusicSymbols
• Rebelo (full, symbols1, and symbols2)
• Fornes Music Symbols (Music_Symbols.zip)
• Capitan (BimodelHandwrittenSymbols.zip)
• Byrd (ByrdOmrTestCorpus.zip)
• Bargheer (MEI Annotations for measures only)
• FreischuetzDigital (MEI Annotations for measures only)
• Baro (BaroMuscima.zip)
• DeepScores V1 with extended vocabulary and selected 100 pages (deep-scores-v1-extended-100pages.zip)
• DeepScores V1 extended (deep-scores-v1-extended.zip)
• DeepScores V2 dense (deep-scores-v2-dense.tar.gz)
• AudioLabs_v1
• Choi Accidentals
• DoReMi V1
• OpenScore Lieder (OpenScore-Lieder-Snapshot-2023-10-30.zip), Snapshot 2023-10-30
• OpenScore String Quartets (OpenScore-StringQuartets-Snapshot-2023-10-30.zip), Snapshot 2023-10-30
• MScoreLib (MScoreLib_all.zip, MScoreLib_Sergey_Prokofiev.zip, MScoreLib_Aleksandr_Scriabin.zip)

Hosting:

• AudioLabs_v2

MuscimaPlusPlusDatasetDownloader now downloads the annotations and the annotated images together, to avoid the need for downloading 2GB if you just need the 7MB of images, that were actually annotated.

This release features a multi-condition version of the CVC-MUSCIMA that consists of files from the Staff-Removal and the Writer-Identification datasets. The images from both datasets have been aligned to have exactly the same dimensions and overlap as good as possible. The dataset contains a total of 10000 png files with 10 different conditions: grayscale, binary, interrupted, kanungo, staffline-thickness-variation-v1, staffline-thickness-variation-v2, staffline-y-variation-v1, staffline-y-variation-v2, typeset-emulation and whitespeckles. Note that these augmented version are just taken from the Staff-Removal dataset and inverted to be consistent with all other images black on white background.