This feature is particularly useful in handwriting and form capture. For instance in numeric fields a handwritten "4" outputs as "A". Another issue is bad scans, where paper artefacts can get detected as punctuation.

Hi @SlappyAUS 👋

I agree but there are two wider topics at hand here:

• the task definition : whether text recognition can use the hypothesis of having no blanks
• handwriting vs typed : handwriting has a difficulty higher than just variability in character style, the spacing and alignments are not fixed. This is a challenge to match with typed ones.

As of right now, we don't support handwriting yet, but we'll have to circle back to this once we do. I suggest checking docTR again once we add handwriting support :)

Also black/white listing is not exactly possible, there are two fundamental options:
whether the blank is a recognized character or not.

so blacklisting is only possible if the model has blank as a recognized character 🤷‍♂️