Dashboard cleaning report

From OTBWiki
Jump to: navigation, search

A reference platform has been settled : ITKv4-Ubuntu10.04-64bits-Debug-BASIC failing tests are investigating below

Basic Filter Tests

tests in fail

bfTvLabelObjectOpeningMuParserFilterTest (Failed)

bfTvLogPolarTransformResample (Failed)


bfTvLabelObjectOpeningMuParserFilterTest (Failed)

-> (input geometry difference updated, test still failing)

bfTvLogPolarTransformResample (Failed) -> (difference due to resampling (10e-5))

problem addressed here : http://scrum.orfeo-toolbox.org/jira/browse/OTB-506

Disparity Map Tests

tests in fail

  • dmTvSubPixelDisparityImageFilterNCC (Failed)
  • dmTvSinusoidDisplacementFieldEstimation (Failed)
  • dmTvCenteredRigidDisplacementFieldEstimation (Failed)
  • dmTvTranslationDisplacementFieldEstimation (Failed)


  • dmTvSubPixelDisparityImageFilterNCC

small difference on metric output when introducing resampling

  • dmTvSinusoidDisplacementFieldEstimation (Failed)

-> difference on many outputs

  • dmTvCenteredRigidDisplacementFieldEstimation (Failed)
  • dmTvTranslationDisplacementFieldEstimation (Failed)

-> small difference in BSpline with Warped* Rescaled* output (no difference if no rescaling )

otb::WarpImageFilter rescaler is a rescale intensity filter

IO Tests

tests in fail

ioTvMultiResolutionReadingInfo_ (JP2 tests) ioTvImageKeywordlistRadarSat1


ioTvMultiResolutionReadingInfo_ (JP2 test) : corrected here http://hg.orfeo-toolbox.org/OTB/rev/d20a22088a1f

ioTvImageKeywordlistRadarSat1 outputs save/load output returns nan values (OTB 3.18 suffers from the same problem)

Failure report

Resampling difference

itkResampleFilter introduce small differences

BSpline Deformation based filter failure

BSpline registration have been modified : [1]

especially this point :

3. The previous version of the B-spline deformable transform set up the grid such that the mesh elements at the upper index of the transform were not used. So, if you wanted to do a 2-D B-spline registration using a mesh size of 10x10, only 9x9 elements would be used in the transform domain. In my opinion this is a bug which should be fixed.

-> problem seems to appear only in some cases :


are successful