B-spline SyN registration has too few control points

[dipterix]

Describe the bug

This error does not display every time, but every now and then it errors out on some images...

Trying to run ants.registration with type of transformation antsRegistrationSyN[b] and encountered error:

Stage 2
  iterations = 100x70x50x20
  convergence threshold = 1.0000e-06
  convergence window size = 10
  number of levels = 4
  using the CC metric (radius = 2, weight = 1.0000e+00, use gradient filter = 0)
  Shrink factors (level 1 out of 4): [8, 8, 8]
  Shrink factors (level 2 out of 4): [4, 4, 4]
  Shrink factors (level 3 out of 4): [2, 2, 2]
  Shrink factors (level 4 out of 4): [1, 1, 1]
  smoothing sigmas per level: [3, 2, 1, 0]
  Using default NONE metricSamplingStrategy 
*** Running B-spline SyN registration (updateMeshSizeAtBaseLevel = [8, 9, 8], totalMeshSizeAtBaseLevel = [4294967295, 4294967295, 4294967295]) ***
Exception caught: 
itk::ExceptionObject (0x112e8d420)
Location: "unknown" 
File: /Users/stnava/code/ANTsPy/itksource/Modules/Filtering/ImageGrid/include/itkBSplineScatteredDataPointSetToImageFilter.hxx
Line: 223
Description: ITK ERROR: BSplineScatteredDataPointSetToImageFilter(0x112ed0ae0): The number of control points must be greater than the spline order.
antsRegistration -d 3 -r [0x152f4fb80,0x142e5e4e0,1] -o [/Users/dipterix/rave_data/raw_dir/UPennOCD3_new/rave-imaging/template_mapping/mni_icbm152_nlin_asym_09a/mni_icbm152_nlin_asym_09a_deformable_,0x29b6f3040,0x29aed55b0] --transform Rigid[0.1] --metric MI[0x152f4fb80,0x142e5e4e0,1,32,Regular,0.25] --convergence [1000x500x250x100,1e-6,10] --shrink-factors 8x4x2x1 --smoothing-sigmas 3x2x1x0vox --transform Affine[0.1] --metric MI[0x152f4fb80,0x142e5e4e0,1,32,Regular,0.25] --convergence [1000x500x250x100,1e-6,10] --shrink-factors 8x4x2x1 --smoothing-sigmas 3x2x1x0vox --transform BSplineSyN[0.1,26,0,3] --metric CC[0x152f4fb80,0x142e5e4e0,1,2] --convergence [100x70x50x20,1e-6,10] --shrink-factors 8x4x2x1 --smoothing-sigmas 3x2x1x0vox -x [0x29ae62c10,NA] --float 1 --write-composite-transform 0 -v 1

Code:

ants.registration(
    fixed = fix_img,
    moving = mov_img,
    mask = mask_img,
    outprefix = outprefix2,
    type_of_transform = "antsRegistrationSyN[b]",
    aff_metric = "CC",
    syn_metric = "CC",
    write_composite_transform = False,
    aff_sampling = 2,
    syn_sampling = 2,
    verbose = True
  )

[ntustison]

I'm guessing this is some system-specific issue. Notice that it's calling the correct command line option with --transform BSplineSyN[0.1,26,0,3] but somehow that 0 in the third parameter is getting translating incorrectly to totalMeshSizeAtBaseLevel = [4294967295, 4294967295, 4294967295].

[dipterix]

4294967295 is $2^{32}-1$. I guess there might be some integer issues...

[ntustison]

Exactly. And considering it's calling the underlying C++ program correctly and I have never produced that, it would be hard for me to debug.

[cookpa]

I've never seen that either but I'll give it a go if given a reproducible example

[dipterix]

What should be totalMeshSizeAtBaseLevel in this case?

[ntustison]

It should be [0,0,0]. That parameter value is by far the most commonly used which is why I'm surprised you're getting it on your machine.

[ntustison]

Closing due to inactivity.