[ITK] ITK bug or feature

[Yaniv, Ziv Rafael (NIH/NLM/LHC) [C]]

Hello Francois,

Tested the patch on gerrit and it fixed the issue (gave +2 on gerrit).

Thanks for the prompt response.
               Ziv

From: Francois Budin <francois.budin at kitware.com>
Date: Monday, September 18, 2017 at 12:06 PM
To: "Yaniv, Ziv Rafael (NIH/NLM/LHC) [C]" <zivrafael.yaniv at nih.gov>
Cc: "community at itk.org" <community at itk.org>
Subject: Re: [ITK] ITK bug or feature

Hello Ziv,
Thanks for the report. There was indeed a bug in itkSimilarity2DTransform and itkCenteredSimilarity2DTransform. I submitted a patch [1] to correct the problem. It would be great if you could test it and review it.
Thanks!
Francois

[1] http://review.source.kitware.com/#/c/22633/

On Thu, Sep 14, 2017 at 12:34 PM, Yaniv, Ziv Rafael (NIH/NLM/LHC) [C] <zivrafael.yaniv at nih.gov<mailto:zivrafael.yaniv at nih.gov>> wrote:
Hello ITK developers,

I encountered unexpected behavior with ITK parameter settings in degenerate cases for the CenteredSimilarity2DTransform and Similarity3DTransform classes.

The degenerate parameter values I am referring to is setting the scale to zero. I encountered this issue when I blindly set all parameters to zero to get an identity transform (obviously scale should be one). It appears that setting the scale factor to zero is acceptable, no exception (arbitrary choice which makes sense for the general SetParameters method – up to the user to make sure the parameter values make sense).

The feature/bug I am referring to has to do with the GetInverse method (ill-defined for this degenerate case):

2D case -  when I print the original transformation, it is identified as singular. When I invoke the GetInverse method vnl generates a warning with the resulting “inverse” transform including Nan and Inf , not identified as singular.

3D case -  when I print the original transformation, it is identified as singular. When I invoke the GetInverse method there is no warning and the “inverse” transform has numbers and is identified as singular.

Is this behavior fine and up to the user to deal with? Should the GetInverse generate an exception?
        thanks
             Ziv

Code for the two cases below:

    {
    typedef itk::Similarity3DTransform<double> TransformType3D;
    // Test degenerate parameter settings
    TransformType3D::Pointer t1 = TransformType3D::New();
    TransformType3D::Pointer t1Inverse = TransformType3D::New();

    // Set parameters
    TransformType3D::ParametersType parameters( t1->GetNumberOfParameters() );
    // Degenerate configuration with zero scale
    parameters[0] = 1.0;
    parameters[1] = 0.0;
    parameters[2] = 0.0;
    parameters[3] = 1.0;
    parameters[4] = 2.0;
    parameters[5] = 3.0;
    parameters[6] = 0.0; //scale

    t1->SetParameters( parameters );
    t1Inverse->SetParameters(parameters);

    t1->GetInverse( t1Inverse );
    std::cout<< "**original**\n"<<t1;
    std::cout<< "**inverse**\n"<<t1Inverse<<"\n";
    }

    {
    // Test degenerate parameter settings
    typedef itk::CenteredSimilarity2DTransform<double> TransformType2D;
    TransformType2D::Pointer t1 = TransformType2D::New();
    TransformType2D::Pointer t1Inverse = TransformType2D::New();

    // Set parameters
    TransformType2D::ParametersType parameters( t1->GetNumberOfParameters() );
    // Degenerate configuration with zero scale
    parameters[0] = 0.0; //scale
    parameters[1] = -21.0 / 180.0 * itk::Math::pi;
    parameters[2] = 12.0;
    parameters[3] = -8.9;
    parameters[4] = 67.8;
    parameters[5] = -0.2;

    t1->SetParameters( parameters );
    t1->GetInverse( t1Inverse );
    std::cout<< "**original**\n"<<t1;
    std::cout<< "**inverse**\n"<<t1Inverse<<"\n";
    }


_______________________________________________
Community mailing list
Community at itk.org<mailto:Community at itk.org>
http://public.kitware.com/mailman/listinfo/community

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/community/attachments/20170919/151b8942/attachment-0001.html>