ENH: Add a parameter to test to improve itk::ResampleImageFilter testing.

[jhlegarreta]

Add a parameter to the itkResampleImageFilterTest2.cxx to improve the
itk::ResampleImageFilter class testing.

Specifically, exercise the workflow when the UseReferenceImage boolean
is set to false. Add the corresponding baselines.

Rename the tests and old baselines to account for this new test.

Take advantage of the commit to improve the style of the test:

• Use the TRY_EXPECT_NO_EXCEPTION in lieu of try/catch for the sake of
  readability/compactness/save typing boilerplate code.
• Explicitly call Update() on filters with the help of the
  TRY_EXPECT_EXCEPTION macro.
• Use the TEST_SET_GET_BOOLEAN macro to test the boolean members.
• Move the std::cout messages before the filter calls to Update().

[jhlegarreta]

Looks like the Off case needs some more work to create a valid target IO region:

itk::ExceptionObject (0x12b0a20)
Location: "unknown" 
File: /home/vsts/work/1/s/Modules/IO/ImageBase/include/itkImageFileWriter.hxx
Line: 286
Description: itk::ERROR: ImageFileWriter(0x12a3820):
Largest possible region does not fully contain requested paste IO regionPaste IO region:
ImageIORegion (0x7fffa282bc30)

The corresponding baselines in the commit are just there to avoid CMake warning that they do not exist (since due to the above exception they were not generated).

However, I'm wondering about a use case where the target region would not be determined by the reference image. Any suggestions? Thanks.

[thewtex]

CC: @romangrothausmann

[romangrothausmann]

Thanks @jhlegarreta for these improvements and thanks to @thewtex for CCing me.

Yes there are use cases, for example if you want to use itkResampleImageFilter for mere down-sizing an image by non-integer scales (i.e. in a way that is not accessible with itkBinShrinkImageFilter). In such a case you just want to specify the desired spacing of the output without providing a reference image, see e.g. https://github.com/romangrothausmann/ITK-CLIs/blob/3153dcd98defe01ccec257336153ed860d494670/resample.cxx#L58-L80.
To my knowledge itkResampleImageFilter is the only available filter in ITK that can be used for such cases (I only learned later, that it is often used with a non-identity transform), which is my motivation to allow streaming at least for such a use case (#82).

[jhlegarreta]

Thanks @romangrothausmann !

I re-worked the test to take the output spacing as an optional parameter in case the reference image is not to be used.

I thought repeatedly at getting rid of the boolean and just using the presence of the reference image as a signal to use it. Having the output spacing also as an optional argument made things less clear to me. If anyone has better suggestion, please go ahead.

Also, I could have made the spacing anisotropic, but the boilerplate code to handle a double-quote enclosed argument holding all spatial dimensions as an array put me off this time.

[jhlegarreta]

CDash reported compiler warnings come from the Examples/DataRepresentation/Image/Image4.cxx file, and are thus unrelated.

The concerned tests are passing:

Line 2855: 1396/2727 Test #1393: itkResampleImageTest2UseRefImageOn 
.....................................................................
Passed    0.04 sec
Line 2859: 1398/2727 Test #1392: itkResampleImageTest2UseRefImageOff
....................................................................
Passed    0.05 sec

[thewtex]

There seems to been some baseline comparison issues with 32-bit systems:

https://open.cdash.org/testDetails.php?test=728564967&build=5731782

[jhlegarreta]

I guess it has to do with this division being rounded differently in a 32-bit system vs. the 64-bit system I use.

Not sure I will be able to test any solution, but I'd guess that doing a ceil would make the test image dimension become 320 at the place of 319. I'll submit a patch set later.

[thewtex]

Fantastic, thanks @jhlegarreta !

[jhlegarreta]

Opened PR #462.