We need a bunch of representative "none" symbols for proper network training.
In a given page, if all (fixed-shape) valid symbols are present in the Annotations structure, we can safely create artificial "none" symbols, based on "empty" areas. But we cannot do so if some valid symbols are missing, because some so-called empty areas may actually contain valid symbols.
The proposal is thus to add this information within the Annotations structure. For example, add an explicit complete attribute like:
<Annotations version="1.0" complete="true">
This would typically apply for all MuseScore generations. It would apply for Audiveris extraction, only after manual validation.
On the opposite (if this complete attribute is false or missing), these annotations would not be safe for artificial "none" symbols creation and will be usable only for their valid symbols.
We need a bunch of representative "none" symbols for proper network training.
In a given page, if all (fixed-shape) valid symbols are present in the Annotations structure, we can safely create artificial "none" symbols, based on "empty" areas. But we cannot do so if some valid symbols are missing, because some so-called empty areas may actually contain valid symbols.
The proposal is thus to add this information within the Annotations structure. For example, add an explicit
completeattribute like:<Annotationsversion="1.0" complete="true">This would typically apply for all MuseScore generations. It would apply for Audiveris extraction, only after manual validation. On the opposite (if this
completeattribute is false or missing), these annotations would not be safe for artificial "none" symbols creation and will be usable only for their valid symbols.