Cloud Vision API - Wrong line order

rsnr · 01-05-2023 01:44 AM

Hi there,

I'm curious how are you handling the too common case where DOCUMENT_TEXT_DETECTION returns the text lines (or specific words) at incorrect order, despite the input being a simple paragraph. For example, line n would jump to line n+2, then backtrack to line n+1.

Using "legacy_layout" does not usually help.

thanks!

anjelab

This is an on-going issue and the engineers are working out a solution for this. For the mean time, you may click +1 on the right pane to express that you are also encountering the same issue.

rsnr

Thanks anjelab. I'm happy to learn the engineers are working on it!

Is there a priority or timeline for the Vision API returning text at the correct order?

There's also a recent, more specific issue, that was opened for this problem here: https://issuetracker.google.com/issues/264322914

While the team is working on solving this issue, the service is not usable for use cases that rely on correct order of the returned text, hence I'm interested in any workarounds that folks may have come up with.