Is it possible to improve drum clef's recognition probability in Audiveris 5.6?

[ligong]

Hello,

I'm currently using Audiveris 5.6 and encountering an issue where drum clefs are not being recognized from scanned images.

Here's an example of the clef I'm referring to:

I was wondering if there are any settings or parameters within Audiveris that can be adjusted to improve its ability to recognize this type of drum clef. Any suggestions or guidance would be greatly appreciated.

Thank you for your help!

[hbitteur]

@ligong
Have you read the documentation about drums?
You need to set a couple of specific flags to allow the recognition of drum staves.

If the percussion clef is still not recognized, you'll have to set it manually.
If so, could you post your input score so that I can see what is happening?

[ligong]

Yes, I followed the recommended drum configuration and achieved good results with other scores that had better image quality.

However, this particular score has a poor resolution. I applied super-resolution to improve it, and while most notes are now recognizable, I'm still having trouble with the percussion clef. I can set it manually to fix it.

The score is attached. Thank you for your help.

[hbitteur]

@ligong
I have studied your example.
Quality is indeed poor, leading to mis-recognized symbols.
But this is not the reason why the 9 percussion clefs were not detected.

The reasons are:

• this percussion clef is very narrow compared with the percussion clefs that have been used to train the glyph classifier.
• the Audiveris support for percussion is very recent, and we haven't yet reached enough training samples.
• we have today 32 samples of percussion clefs (vs 2314 for treble clefs, 2217 for bass clefs, ...)
• moreover, among the 32 samples, 30 are wide and only 2 are as narrow as yours.

The only way to improve this situation is to provide more representative samples of percussion clefs.
Perhaps you could point me to available collections?

[ligong]

Thank you for helping to clarify the root cause. While I currently only have this single poor score image, I'll be sure to provide additional samples here if I encounter them. This project is truly impressive, and I appreciate your efforts