OCR language/page segmentation, Google Cloud Vision and FUNSD format

naourass commented 2 years ago

Hello and many thanks for this awesome tool.

For the needs of my current use case, I added these features to my local pawls clone :

OCR argument for choosing tesseract language and segmentation type
Google Cloud Vision OCR support which may be more accurate than tesseract in some cases (should download credential file from GCP to root folder)
Command to generate FUNSD-like json with labels, text and words from coco and tokens csv.

If these features or any of them is relevant to pawls vision, let me know so I can make a PR.

lolipopshock commented 2 years ago

Sounds great -- looking forward to your PR!

naourass commented 2 years ago

Hi there, I've just made a first PR. The second one will be following soon with FUNSD-like and IOB export. Let me know if you have any remarks !

allenai / pawls