invalid floating point operation when calling TessBaseAPIAnalyseLayout

[Matthias Schneider]

Hi,

I'm trying to use the C-API to detect text regions on checks.
During my first tests, I ran into an 'invalid floating point operation' happening on just one of my sample images (testcheck.png).

This is my (simplified) test application:

  
    TessBaseAPI* hTesseract = TessBaseAPICreate();
   
    TessBaseAPIInit2(hTesseract, "", "eng", OEM_DEFAULT);
  
    PIX *img = pixRead("testcheck.png");
    TessBaseAPISetImage2(hTesseract, img);
    TessBaseAPISetPageSegMode(hTesseract, PSM_AUTO);
    TessBaseAPIAnalyseLayout(hTesseract);    // here I get the 'invalid floating point operation'

Is this a bug or am I doing something wrong?
As I already said, this happens only one some of my sample images while it works fine on others.

Matthias

[Tom Morris]

What version of Tesseract? Source or binary? If source, compiled with what compiler & compilation options? What operating system? What processor architecture? What's the stack trace for the crash?

[Matthias Schneider]

I'm using latest dev version 3.05.00dev and I used peirick/leptonica (https://github.com/peirick/leptonica) to build libtesseract.dll and liblept.dll with Visual Studio 2015.
However, the resulting DLLs I'm using in Embarcadero C++ Builder 10.1 after I created the import libs with implib. And after doing some more debugging it looks like I found the reason: in baselinedetect.cpp there is this code:

  for (int i = 0; i < positions.size(); ++i)
     offsets.push_back(fmod(positions[i], *m_out));

In some cases *m_out can be 0.0 causing fmod to return 'nan' on Visual Studio projects but causing an 'floating point invalid operation' on C++Builder projects.

[Tom Morris]

On Monday, June 13, 2016 at 10:22:32 AM UTC-4, Matthias Schneider wrote:

I'm using latest dev version 3.05.00dev and I used peirick/leptonica (https://github.com/peirick/leptonica) to build libtesseract.dll and liblept.dll with Visual Studio 2015.
However, the resulting DLLs I'm using in Embarcadero C++ Builder 10.1 after I created the import libs with implib. And after doing some more debugging it looks like I found the reason: in baselinedetect.cpp there is this code:

  for (int i = 0; i < positions.size(); ++i)
     offsets.push_back(fmod(positions[i], *m_out));

So, this line https://github.com/tesseract-ocr/tesseract/blob/master/textord/baselinedetect.cpp#L761 

 

In some cases *m_out can be 0.0 causing fmod to return 'nan' on Visual Studio projects but causing an 'floating point invalid operation' on C++Builder projects.

Have you tried switching the floating point processing to be compatible in your target compiler?

This looks like it could be relevant: http://docs.embarcadero.com/products/rad_studio/delphiAndcpp2009/HelpUpdate2/EN/html/devcommon/compdirsfinitefloat_xml.html

ie the {$FINITEFLOAT OFF} directive

Tom