GSoC 2018 - Machine Learning Dataset for OMR - Week 12

Posted 6 years ago

Heya!
Worked on testing the entire set once again. Finalised on some issues which I faced during testing. For the future, I will be working on those issues. Hope we take this project to new heights. Coming to this week's analysis,

Below is the segmented status of the project:

Current status of the project
We are done with:
1. Porting the OMR work from imeta to master.
2. Grace Notes Implementation for OMR tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/27
3. Bracket Implementation for OMR tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/26
4. Tuplet Implementation for OMR tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/22
5. Time Signature Upper and Lower Halves annotation for OMR tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/28
6. Rest Dot Implementation for OMR tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/23
7. Simple Image URL for OMR tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/30
8. Staccato Dot for OMR tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/25
9. SMuFL symbols identifier for OMR tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/29
10. Repeat dot implementation for OMR tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/24
11. Crash error rectification while XML generation. - An issue which was faced while testing it on different kind of scores.
12. Testing of scores on Musescore so that they generate XML. The application has been tested on a dataset of 988 scores and it works perfectly.
13. Tuplet Implementation made better.
14. Grace Note Implementation corrected. It now has the nested approach as discussed in the issue: https://github.com/Audiveris/omr-dataset-tools/issues/27. Some samples can be seen in the comments.
Clef Implementation made better. No Sym issue corrected. Made changes for using SMuFL names in the code.
Ported all these changes to nasehim7/imeta, which is rebased with the nasehim7/2.3 which has the latest changes from MuseScore/2.3, at this moment.
https://github.com/nasehim7/MuseScore/compare/2.3...nasehim7:imeta
15. Image Format changed to support grayscale image tackling the issue: https://github.com/Audiveris/omr-dataset-tools/issues/31. Testing and resolving issues that I came across. Giving more structure to the code I wrote before. Some changes to my previous commits and adding those to imeta.
16. Initialized segmented testing of our test data set in chunks of 150 to 200 depending on the complexity of the scores. First, 165 scores are done.
17. Done testing of 350 more scores and found an issue. Worked on it. Found some missing shapes.
18. Done with testing the remaining 472 scores. Some issues encountered and refined many of our implementation for better(mostly our nasehim7/imeta). Some shape additions again.
19. Resolved previous week issues.

Added: Tested the entire once again to check and jot down issues. Refining. I have noted all of them and now I will be working on them. Made the imeta compatible with 2.3.2 branch as discussed.

Key accomplishments this week
Testing the entire dataset again. Noting the issues. Refining. imeta rebased with 2.3.2 and created a new PR.

Key tasks that stalled
None

Tasks in the upcoming week:
Work on the noted issues one by one. Also, get our imeta branch working with AppVeyor.

I will be posting for 2 weeks more as discussed with lasconic. Lots and lots of work is needed but yes we have things structured now with the work we have done in GSoC 2018. GSoC is amazing, Musescore is amazing, this whole scenario of being able to contribute to some of the best projects in the world is amazing.

Best,
Animesh
Github: https://github.com/nasehim7