davanstrien maps OCR models across four Hub sub-collections

davanstrien's OCR on the Hub collection groups OCR models on Hugging Face into four sub-collections: 17 for documents, 10 for languages and scripts, 10 for handwriting and archives, and 6 for text recognition and pipelines. Two experts in our Who's Who directory shared the collection.

The framing is deliberately practical. 'Curated OCR models for documents, languages, handwriting and text in images,' the overview reads. 'Browse four collections with short practical notes.'

The languages list covers Thai, Japanese, Vietnamese, Arabic, Korean and Devanagari. The handwriting list spans Swedish, Norwegian, German Kurrent, Tibetan and Hebrew-script manuscripts. The pipelines section features Kraken and PaddleOCR for documents, manga and text in photographs. Contributions came from stefan-it, maehr and sharonshorg.

Each sub-collection ships with a caveat. The documents note warns that 'some models need a full parsing pipeline.' The handwriting note is blunter: 'Many models need cropped lines.' The languages note tells users to 'Check each model's expected input and domain.'