[FEATURE REQUEST] Model Fine-Tuning

Is your feature request related to a problem? Please describe.
Better document recognition. The auto-cropping of isolated documents work well, but the cropping of e.g. single pages of a book need some help.

Describe the solution you'd like
Implement a fine-tuning loop based on the usage of the app.

Thousands of people manually adjust / improve the cropping region on their phones.
That could be used as fine-tune training data for further improvement.

Probably the easiest way is to add a toggle option for data donations in the settings and collect the data.
Better but I don't know if possible: Do it on the users phone and send back some anonymous data without sensitive informations.

Describe alternatives you've considered

Additional context

P. S. Thank you for the already amazing work!
P. P. S. Or are there already existing cropping data sets for book pages?

The issue names no files, tests, or entry points. Start by clarifying the data-donation and fine-tuning scope with maintainers; done would require an agreed design covering data collection, privacy, and the training workflow.