ITK Python Image of itk.D Vectors For Composite Transform

josebouza · June 21, 2022, 9:32pm

Hello,

I am trying to convert a numpy array to an image of type itk.Image[itk.Vector[itk.D,3],3] in python ITK. The issue is that most ITK filters do not support this type. I would rather avoid having to compile my own ITK version, so I am trying to figure out if there is any workaround. Here are some things I’ve tried:

Use itk.image_from_array to load numpy array into itk.VectorImage type, and then use CastImageFilter to convert this to itk.Image[itk.Vector[itk.D,3],3]. Python CastImageFilter does not support this output type.
Use itk.image_from_array to load numpy array into three different itk.Image[itk.D,3] images and then combine them with itk.ComposeImageFilter. Again, itk.ComposeImageFilter does not support itk.Image[itk.Vector[itk.D,3],3] as an output type.

The reason I need itk.Image[itk.Vector[itk.D,3],3] is to initialize a itk.DisplacementFieldTransform object, which only appears to support images of vectors (not itk.VectorImage). Any help would be greatly appreciated!

dzenanz · June 27, 2022, 12:07pm

Have you tried casting your numpy array into 32-bit floats, then using itk.Image[itk.Vector[itk.F,3],3]? There is usually no need for such high precision for displacement fields, hence it is not pre-built.

josebouza · June 27, 2022, 1:50pm

Thank you for the reply. The reason for wanting a displacement field of type itk.Image[itk.Vector[itk.D,3],3] is that I eventually want to use the displacement field in a CompositeTransform and itk.CompositeTransform only exposes the template class for type itk.D.

Maybe a future ITK version should expose itk.CompositeTransform for itk.F to make this use case easier.

dzenanz · June 27, 2022, 1:59pm

@Tom_Birdsong @Pranjal_Sahu did you run into this issue, and if so how did you solve it?

Pranjal_Sahu · June 27, 2022, 2:03pm

Please check this test file.

github.com

InsightSoftwareConsortium/ITK/blob/master/Modules/Filtering/DisplacementField/wrapping/test/itkDisplacementFieldTransformTest.py

# ==========================================================================
#
#   Copyright NumFOCUS
#
#   Licensed under the Apache License, Version 2.0 (the "License");
#   you may not use this file except in compliance with the License.
#   You may obtain a copy of the License at
#
#          https://www.apache.org/licenses/LICENSE-2.0.txt
#
#   Unless required by applicable law or agreed to in writing, software
#   distributed under the License is distributed on an "AS IS" BASIS,
#   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#   See the License for the specific language governing permissions and
#   limitations under the License.
#
# ==========================================================================*/
import itk
itk.auto_progress(2)

This file has been truncated. show original

dzenanz · June 27, 2022, 2:57pm

This test does not address composite transform use of DF transform.

Anton_Naumov · February 2, 2024, 12:12pm

I ran into exactly the same issue – one can define the DisplacementField transform of a float type, but cannot add it to a composite transform since it can only be of a double precision.

@Pranjal_Sahu, I’ve checked your file, but it doesn’t cover the CompositeTransform case.

Any improvements or ideas? The only workaround I’ve found is to cast dvf through numpy, but it is ugly

matt.mccormick · February 2, 2024, 6:59pm

Hi,

We can also wrap Composite Transform for itk.F – here is a patch: