Error when doing both detection and recognition on image about tamil_ocr HOT 10 CLOSED

Unfortunately I cannot share example of images, but I can say that when no readable (low res) text is present the code fails in this way. For now I'm handling it with a try except and print in the except ' no readable text found'

from tamil_ocr.

Thanks for reporting this @Dario-Mantegazza , I will handle this in next version of the code.

Quick hack that may help in the low resolution text
Lower the threshold for text_threshold (can accept any value between 0 to 1). This can be done as below when initializing the ocr module.
ocr = OCR(detect=True,text_threshold=0.2)

from tamil_ocr.

@Dario-Mantegazza , Please update ocr_tamil package using below command.

pip install -U ocr_tamil

from tamil_ocr.

Thanks for the quick reply, I will also try to lower that threshold, (OT, it would be ideal to have a rough guide/doc on the different params and their effect).
Regarding updating the package, I'm now rebuilding my container image and I will update you on the results :)

from tamil_ocr.

@Dario-Mantegazza , Couldn't able to reproduce the issue. I have run the inference on the image you have shared above and attached the colab notebook that I have used to run this.

https://colab.research.google.com/drive/11tMrKEeMtoKILwFP8TeChnI8_bEIeulA?usp=sharing

OK, I will check if it is a problem with my setup and let you know next week. Thanks for your support in the meantime, it was super

from tamil_ocr.

sorry I got some unforeseen delays, I will probably get back to you next week

from tamil_ocr.

Ok it works. Earlier I was probably working with some cached version of the lib. My setup is not straightforward and this could have happened. Thanks for fixing the bug. If I have more issues like this I will post them here, but I'm sure this will not happen :)

from tamil_ocr.

I'm getting the same error. The container build, installed the version 0.3.6, I don't know if this helps.
At least by changing the threshold, now more images are read without errors.
Edit: I think I have a picture that I can share that fails even with the low threshold. here it is clear that no text is readable, this sometimes happens in our pipeline.

from tamil_ocr.

@Dario-Mantegazza , Couldn't able to reproduce the issue. I have run the inference on the image you have shared above and attached the colab notebook that I have used to run this.

https://colab.research.google.com/drive/11tMrKEeMtoKILwFP8TeChnI8_bEIeulA?usp=sharing

from tamil_ocr.

from tamil_ocr.