Text Detection and Document Text Detection (OCR) - Different results with the same file (Local vs Cloud)

Question

I'm having a problem with the new version of Vision API Text Detection and Document Text Detection (OCR). Version that will become default from February 19, 2024.

I am using the same project and the same file but in different locations.

When I run it locally I get one result (the correct one), but when I run it on an Azure VM I get the result with some differences (the wrong one).

I didn't find anything in the documentation about any environment configuration influencing the result.

I'm using ImageAnnotatorClient/AnnotateImageRequest.

Builder builder = Feature.newBuilder();
builder.setModel("builtin/latest");
builder.setType(typeDetect); (Feature.Type.TEXT_DETECTION or Feature.Type.DOCUMENT_TEXT_DETECTION)
Feature feat = builder.build();

When I use the current builtin/stable (default) version, I don't have these problems.

The errors are as follows:

Local(Correct): -1.898,37
VM(Wrong): -1.898**.**37

Local(Correct): 2.340,772 24.703,952 30
VM(Wrong): 2.340,772 24.703,952 w

Local(Correct): 2.290,092 23.857,148 31
WM(Wrong): 2.290,092 23.857,148 w

I tried the following solutions which didn't work:

1. Send "Hints".
   
2. Set the type as DOCUMENT_TEXT_DETECTION
   
3. "Process only in the USA"