@jbvimort I was about to bump the latest version of the ITKTextureFeatures when I realized that it has not been added as a remote to ITK yet.

I'm wondering whether it should be added as a remote to ITK, or whether you are waiting for #3 to be tackled.

May be Matt, Beatriz and/or François should may be inquired to be sure?

Description

[4696/5195] Generating ../../Typedefs/itkCoocurrenceTextureFeaturesIma...eFilter.idx, ../../Typedefs/itkRunLengthTextureFeaturesImageFilter.id
itkCoocurrenceTextureFeaturesImageFilter: warning(4): ITK type not wrapped, or currently not known: itk::ImageIOBase
itkRunLengthTextureFeaturesImageFilter: warning(4): ITK type not wrapped, or currently not known: itk::ImageIOBase

Expected behavior

All classes wrapped.

Reproducibility

Wrapping failure.

Versions

2020-03-18 master branch.

Hi. I am getting seg fault when I try to execute the code bellow. Any idea why?

import itk
import numpy as np

img = np.random.randint(0,256,64**3).reshape(64,64,64).astype(np.uint8)

msk = 1 + np.zeros_like(img)

img = itk.GetImageFromArray(img)
msk = itk.GetImageFromArray(msk)

filtr = itk.CoocurrenceTextureFeaturesImageFilter.New(img)
filtr.SetMaskImage(msk)
filtr.SetHistogramMinimum(0)
filtr.SetHistogramMaximum(255)
filtr.SetNeighborhoodRadius([3,3,3])
filtr.SetNumberOfBinsPerAxis(4)
result = filtr.GetOutput()

I am running with

python 3.6
itk                      5.0rc2     
itk-core                 5.0rc2     
itk-filtering            5.0rc2     
itk-io                   5.0rc2     
itk-numerics             5.0rc2     
itk-registration         5.0rc2     
itk-segmentation         5.0rc2     
itk-texturefeatures      3.2.1      

When using an image with signed pixel type as input, the output is wrong.
Minimal example:

#include "itkImage.h"
#include "itkVectorImage.h"
#include "itkCoocurrenceTextureFeaturesImageFilter.h"

template<typename TImage>
void run()
{
	typename TImage::Pointer image = TImage::New();
	typename TImage::RegionType region;
	region.SetIndex({{0, 0}});
	region.SetSize({{5, 5}});
	image->SetRegions(region);
	image->Allocate(true);
	
	using TextureType = itk::Statistics::CoocurrenceTextureFeaturesImageFilter<TImage, itk::VectorImage<float, 2>>;
	typename TextureType::Pointer texture = TextureType::New();
	texture->SetInput(image);
	texture->Update();
	
	std::cout << texture->GetOutput()->GetPixel({{2, 2}}) << std::endl;;
}

int main()
{
	run<itk::Image<signed int, 2>>();
	run<itk::Image<unsigned int, 2>>();
}

Output:

[0, 0, 0, 0, 0, 0, 0, 0]
[1, 0, 0, 1, 0, 0, 0, -0.00392157]

The problem comes from the DigitizerFunctor, on line 80 the computing of the output range overflows.

Having ITKTextureFeatures available through conda would be much appreciated.

Python 3.9.4

pip install itk -U
pip install itk-texturefeatures

ITK 5.2.0 python install + itk-texturefeatures 3.5.0 install.

help(itk.Image)

results in the following warning:

itkFixedArrayF8 not loaded from module TextureFeatures because of exception:
 module 'itk.TextureFeaturesPython' has no attribute 'itkFixedArrayF8'
itkFixedArrayF10 not loaded from module TextureFeatures  because of exception:
 module 'itk.TextureFeaturesPython' has no attribute 'itkFixedArrayF10'

The following are missing best practices:

• Add MaskImageType definition, and pixel type
• Use Macros for setting named input
• Make pipeline input named, number and optional

See 95705f5 for similar changes.

The occasional test segfaults and inconsistency hopefully can be addressed by taking care of some memory access bugs identified by valgrind. These can be obtained by:

1. Install valgrind
2. Build ITKTextureFeatures with CMAKE_BUILD_TYPE set to RelWithDebInfo
3. Run ctest -D ExperimentalMemCheck -R RunLengthTextureFeaturesImageFilterInstantiationTest -VV, etc. Or exclude -R to run all the tests.
4. Address the issues identified in Testing/Temporary/MemoryChecker.*.log

256 makes computation very slow, and this is larger than the typical number of effective intensity bins encountered in many images. A better default is 8, 10, 16?

I am not sure that's correct, also there is no intensity range variable to tweak that value.

Also here:

A starting point:

https://github.com/InsightSoftwareConsortium/ITKPerformanceBenchmarking/blob/master/README.rst

It should have:

• Title with the name of the repository
• Badges for AppVeyor, CircleCI, TravisCI
• Short description
• Pointer to IJ article
• Pointer to Slicer Extension
• Pointer to Python packages
• Description of how to enable in ITK
• Funding acknowledgements

For some inputs, nan values can appear in the output. From the few tests I did, it seems to happen with floating values and isolated points in the mask.

Minimal example:

#include "itkImage.h"
#include "itkVectorImage.h"
#include "itkCoocurrenceTextureFeaturesImageFilter.h"

typedef itk::Image<float, 2> ImageType;
typedef itk::Image<unsigned char, 2> MaskType;
typedef itk::VectorImage<float, 2> VectorType;
typedef itk::Statistics::CoocurrenceTextureFeaturesImageFilter<ImageType, VectorType> TextureType;

template<typename TImage>
typename TImage::Pointer createImage()
{
	typename TImage::Pointer image = TImage::New();
	typename TImage::RegionType region;
	region.SetIndex({{0, 0}});
	region.SetSize({{5, 5}});
	image->SetRegions(region);
	image->Allocate(true);
	return image;
}

int main()
{
	ImageType::Pointer image = createImage<ImageType>();
	MaskType::Pointer mask = createImage<MaskType>();
	mask->SetPixel({{2, 2}}, 1);
	TextureType::Pointer texture = TextureType::New();
	texture->SetInput(image);
	texture->SetMaskImage(mask);
	texture->Update();
	
	std::cout << texture->GetOutput()->GetPixel({{2, 2}}) << std::endl;
}

Output:

[nan, 0, nan, nan, nan, nan, nan, nan]

xref #33

ref @phcerdan idea's of using new iterators in ITK5

Nice work here, I look forward to the insight journal article.

Just wanted to see if there was any interest in a filter which computed first order statistics (mean, min, max, variance, sigma, skewness, kurtosis, and entropy) over a sliding window of a scalar image. They seem very much inline with the subject of this repo.

I have an implementation here:
https://github.com/blowekamp/itkTextureAnalysis/blob/master/include/itkTextureMovingHistogramImageFilter.h
https://github.com/blowekamp/itkTextureAnalysis/blob/master/include/itkTextureHistogram.h

If you are interested an I make a pull request with these classes with tests and cleanup ( create? ) documentation for them.

The ITK remote module is a bit out of date now pointing to 2fed5dc.

Test the example build on CircleCI 2

Traceback (most recent call last):
  File "/home/user/a.py", line 27, in <module>
    filtr = itk.CoocurrenceTextureFeaturesImageFilter.New(im)
  File "/home/user/py_env/lib/python3.7/site-packages/itk/support/lazy.py", line 76, in __getattribute__
    base.itk_load_swig_module(module, namespace)
  File "/home/user/py_env/lib/python3.7/site-packages/itk/support/base.py", line 110, in itk_load_swig_module
    l_module = loader.load(swig_module_name)
  File "/home/user/py_env/lib/python3.7/site-packages/itk/support/base.py", line 259, in load
    l_spec.loader.exec_module(l_module)  # pytype: disable=attribute-error
  File "<frozen importlib._bootstrap_external>", line 728, in exec_module
  File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
  File "/home/user/py_env/lib/python3.7/site-packages/itk/support/../TextureFeaturesPython.py", line 74, in <module>
    from itk.itkCoocurrenceTextureFeaturesImageFilterPython import *
  File "/home/user/py_env/lib/python3.7/site-packages/itk/itkCoocurrenceTextureFeaturesImageFilterPython.py", line 3563, in <module>
    from itk.support import itkHelpers
ImportError: cannot import name 'itkHelpers' from 'itk.support' (unknown location)

Just updated itk package from pip to version 5.2 and I cannot run itk-texturefeatures anymore, here it's the Traceback for the example code found on this repository.

Searching a bit, I found this commit from ITK repo which deleted filter_args(object) from itk.support.helpers which also is referenced on itkCoocurrenceTextureFeaturesImageFilterPython.py as from itk.support import itkHelpers. So it was renamed from itkHelpers to helpers.

Also, accept_numpy_array_like_xarray() was changed to accept_array_like_xarray_torch()

helpers.py source file

Hope this information somehow help.

CC: @csheaff

With the merge of #41, the C++ interface has changed the default for the mask image type. Currently, they Wrapping still specifies the mask image type to be the same as the input images. Should the wrapping be changed?

Also consider wrapping for "VectorImage" as can simplify the usage, by not needing to specify the image vector size as part of the image type.