search

mindee / doctr

docTR (Document Text Recognition) - a seamless, high-performing & accessible library for OCR-related tasks powered by Deep Learning.
https://mindee.github.io/doctr/
Apache License 2.0
3.83k stars 435 forks source link

Support for detecting special symbols like bullet points, check boxes, etc. #1570

Open parthpatel002 opened 6 months ago

parthpatel002 commented 6 months ago

🚀 The feature

Currently, all text detection models in docTR do not seem to identify special characters like bullet points, check boxes, etc. (likely because the training data is so) - attaching sample outputs (clipped screenshots of document pages). We should be able to detect these symbols to increase detection coverage to all text present on a page. Bullet points:

image

Checkboxes:

image

Motivation, pitch

Symbols like bullet points, checkboxes, etc. form an integral part of the text content of many types of documents in general and OCR should be able to detect as well as recognize these symbols to increase coverage to all text present on the page.

Alternatives

No response

Additional context

No response

felixdittrich92 commented 6 months ago

Hi @parthpatel002 :wave:,

with bullet points i agree (but as you already mentioned there are no such samples in the pretraining dataset). About the check boxes i think this would be more a topic for document layout parsing :)

But we have on the roadmap to pretrain on https://huggingface.co/datasets/pixparse/pdfa-eng-wds