OMR doesn't work

• Sep 7, 2011 - 12:44
S5 - Suggestion

I saw a showreel about OMR in MuseScore 2...can I ask why doest work now? OMR is most biggest upgrade of MuseScore...very important to fix that.

Attachment Size
error.jpg 16.14 KB


There is little OMR in MuseScore 2.0 and it works only on very particular PDF files, Black and white, they need preprocess before. Moreover, the OMR features in MuseScore only detect staves and measures for now. Not the notes. It's already a hard job, and can save a lot of time for transcription, but don't expect much.
There is no crash so the "bug" is far from critical, plus without the file, we can't figure out what the unknow error is.

It depends, there is another bug report related to this. Musecore crash while opening a pdf file.
I tested this on a personnal score scanned in gray levels.
If you need some more informations, I can post a gdb log of the crash (and the pdf score too).


The program doesnt crash but shows warning message than cannot open this file...I use Windows 7 64bit and clean printed score from Sibelius as pdf file...It couldnt be nessesery read notes, but staffs (1-6 lines) and maybe bars...audiverse 3.3 cannot read big score and in audiverse 4beta cannot manage settings (in program, not in java settings)and PhotoScore or Capella Scan is best solution for OMR but its too expensive and cant upgrade programs meaningfully...

Under linux, I converted the gray pdf into black and white using:

gs -sDEVICE=psmono -dNOPAUSE -dBATCH -dSAFER Input.pdf
ps2pdf Output.pd

And now my file is opened correctly in musecore.
As specified in a previous comment, I've only got a bars and no notes.


@ycollet Indeed, only bars and no notes, that's how it is supposed to work. Werner decided not to pursue anything beyond recognizing the systems, staves and bar lines. Currently it works best for piano sheet music. So basically, this OMR feature in MuseScore will help you to transcribe/copy faster, but that's all.

That is good reason to make little OMR...
1. Today OMR for printed score is not so bad but, if you want to recognize it, it will be to globaly 45-95 percent accurate (dependent to style of score - old pocket score or new printed score)
2. OMR for handwrite is more than poor...hope for learning system for your own handwrite style is only way to do it or AI to recognize specific symbols and evaluation of them (as for 1st poit)
3.If you use this both poits above now...better way is manually set notes to staffs...less time to do, abolition of time occupied repairing recognized score and avoid to error in import...

If you want to make real OMR it will be long time to do it...good luck

For more full-featured OMR, there's the free Audiveris program, written in Java. It even has a plugin to open scores in MuseScore.

Yes indeed. I haven't checked its progress lately, but last time I looked results were still very poor compared to commercial programs like Sharpeye.