#ocr4all — Public Fediverse posts

Every now & then, I give #ChatGPT a scan of my handwriting to test its skills in working with #handwrittentexts. Initially, it responded that it could not process the scans or gave me entirely fictional output, but today it got almost everything right. These results are better than those I achieved with #HWR models in #Tesseract & #OCR4all without additional training. I also asked ChatGPT what it "thought" about my writing & it called it "consistently shaped & large with stylistic strokes."

@tkinias as far as I understand you want to implement a PDF -> Text -> PDF workflow. Using plaintext as intermediate is problematic, as you (may) lose a lot of layout information.

For high quality fulltext you may need a more sophisticated intermediate format like #PageXML or #AltoXML. But they also require a more sophisticated tool for editing like #OCR4All.

Salut ici :)
Je suis en train de tester #ocr4all pour faire reconnaître de l’écriture manuscrite. ( #ocr #hwr #htr )
Mais j’arrive à rien.
C’est peut-être à cause des modèles ?! Je n’ai que ceux de base qui sont optimisé pour le vieux français … ça aide pas … 😅

Est-ce que quelqu’un a déjà essayé et réussi ??

#question #RT apprécié 😌

#Day2 of #DH2023 pre-conference workshops. Today I am learning how to use #OCR4All. Hopefully, I can teach and tutor folks at the #UniversityOfOslo later. It could be especially useful for #MedievalManuscripts since we have a couple of projects that require good #OCR #HTR processing!