OCR on image

#28
by glitchyordis - opened

Obtaining key information is quite straightforward but Is there a way to obtain bbox locations from texts detected?

glitchyordis changed discussion title from OCR text to OCR on image

You can prompt the model to return bbox locations (see here: https://huggingface.co/spaces/maxiw/Qwen2-VL-Detection). I also tried "detect all texts" but the results are not super precise.

I tried OCR on a not-that-clear text screenshot, it's working nearly perfectly. But the model seems not good at recognize twisted text. E.g. words on bottle.

This comment has been hidden

Has anyone managed to get OCR text detections and their corresponding bounding boxes using QWEN2-VL-7B-Instruct model accurately? I am able to get OCR detections correctly but not the boxes. The boxes are quite misplaced and random I'd say.

Has anyone managed to get OCR text detections and their corresponding bounding boxes using QWEN2-VL-7B-Instruct model accurately? I am able to get OCR detections correctly but not the boxes. The boxes are quite misplaced and random I'd say.

what's your settings especially your system prompts? I wonder how to get the ocr text without bbox.Thanks

Sign up or log in to comment