I am having issues with the language setting in Google Vision OCR within the workflows. In some cases, the language is automatically set to “und,” which results in inaccurate outputs. I would like to explicitly specify the language to ensure better accuracy, specifically setting the language to “en”. Is it possible to do this in the workflows?
Hi @IA_Prosys ,
Thank you for sharing suggestion!
Can you share screenshot of outputs when google OCR worked correctly?
Thanks,
Grzegorz
Hi @IA_Prosys,
With next release of inference you will be able to specify a list of language hints that are passed to google OCR API. This will give you ability to experiment with this parameter.
Hope this helps,
Grzegorz
I’m testing the new functionality, but it seems it’s not properly recognizing the language input. What I don’t understand is why with ‘und’ it appears to read things backwards, though this only happens in some cases. Has anyone else experienced this?
Hi @IA_Prosys,
In terms of Google OCR API, documentation suggests API should take hint under account, unfortunately there is not much more information available that would explain behavior you report. You can play with API directly by making requests to this endpoint
Hope this helps,
Grzegorz
This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.