Google Cloud vision demo behaving differently from actual API?

Question

The same image leads to different text detection results in the google cloud vision API demo versus the actual API. In the demo, the accuracy is much higher. More importantly, the newline behavior is more correct in the demo; blocks of text are treated as together, whereas in the API I'm using with the free trial, the ordering of the text is treated as strictly "top to bottom" with no regard for horizontal proximity. Am I doing something wrong, or is this a bug?

Answer 1

It's possible that you may get a better result from detect_document_text than from detect_text. Try both APIs.

Answer 2

I'd comment but I can't:

@class is right: I had the same issue with different results from the demo and the API. Changing the API request to use DOCUMENT_TEXT_DETECTION instead of TEXT_DETECTION gave comparable results in the API and demo.