OMR Bootstrapping

To get beyond the synthetic and the deteriorated-synthetic score images, we can think of a smart use of Audiveris OMR.

Once Audiveris V5 gets sufficiently efficient (thanks to the training on synthetic images) we can consider that its output on "real" scores will get better, leading to the production of more reliable detected symbols.

Using an interactive tool, such as the vintage tool of Nicolas, we could ask a human being for validation. And validated Audiveris OMR data easily leads to the production of reliable annotations.

If the amount of validation work remains low, this is a sizable way to progressively integrate real scores material into the OMR dataset.