Question on the lines when detecting and recognizing text.

[HuSSonO3]

Hayoo!,

I am trying to read a document where the horizontal part of it is all connected.
for exp:

i want the model to assume that the total with the number is the whole line
but the lines that get recognized are very close words only not far apart ones, is there a way to do that?
Maybe just look for every box on the same x level and group it as a line.

[felixdittrich92]

Hi @HuSSonO3 👋,

You can pass a higher value for paragraph_break to the ocr_predictor the default is 0.035 so if you pass for example 1 instead it should merge every horizontal in one line.

Example:

model = ocr_predictor(pretrained=True, paragraph_break=1)

Hope this helps :)

Best,
Felix

[HuSSonO3]

Thank you so much!.