Amount of returned tokens is changing from document_text_detection

[Iason Georgakopoulos]

I have a test that runs that checks the tokens extracted for a given page. This is set up since October and in the last 24hours the amount of tokens returned for the same given pages seems to be changing. The change is between 1 or 6 tokens, but it has changed multiple times. Is Google retraining or updating the model? Are you aware of a similar behaviour?

[Iason Georgakopoulos]

Seems to be the case that Google is upgrading the models. We found out that for pages recognised as GERMAN (language) no more arabic characters are recognised. That was previously the case.

[Sujoy Basu]

Please open a ticket in the private issue tracker,  attach image(s) where your test fails, and someone from engineering team will investigate. Thanks!