Add streaming base classes and streaming statistics by blowekamp · Pull Request #855 · InsightSoftwareConsortium/ITK

blowekamp · 2019-05-10T19:44:12Z

No description provided.

blowekamp · 2019-05-13T14:02:29Z

Changing the base class for the StatisticsFilters to the ImageSink with streaming support when very smoothly. More testing and documentation passes are still needed.

Please review the style interface for the StreamingProcessObject and ImageSink classes.

A backwards incompatibility change introduced is that the LabelStatisticsImageFilter and StatisticsImageFilter no longer produce an image as output since they are no longer derived from ImageToImageFilter.

dzenanz

Looks good.

dzenanz · 2019-05-14T19:50:47Z

Looking good so far.

blowekamp · 2019-05-14T20:42:59Z

I have completed refactoring the classes. I'll write some detailed/corner cases tomorrow to wrap things up.

@romangrothausmann Please check out this PR and test the classes. Thank you!

zivy

Some comments in line. Still a review in process (will continue tomorrow).

romangrothausmann · 2019-05-15T09:50:19Z

Many thanks @blowekamp, this is a very great contribution! Your former module for ITK4 has helped us a lot when we were in the need to get statistics of very large datasets and we worried about its missing compatibility with the transition to ITK-5. With this PR this would be solved and its future maintenance ensured. In addition it is great that this PR also includes the ideas of blowekamp/itkStreamingSinc#3.

zivy

Added some more comments / questions. Nothing that is too critical.

The StreamingProcesObject provide a pipeline mechanics to update the input's pipeline with a sequence of regions and generically execute them in a StreamedGenerateData method.

Adding the ImageSink base class which provides support for streaming and multi-threading on an input image. Generally this is used when transforming an image into something like computing statistics, or a RLE of an image.

Add functionality, to compute the label statistics of labeled image in a streamed or out-of-core way. For example, from a very large segmented image the bounding boxes can be computed to determine a region of interest to load.

And the same change from ImageToImgeFilter to verify that input images are congruent in physical space by default.

Change the base class from ImageTransformer to ImageSink and expose SetNumberOfStreamDivisions method to enable streamed generation of Histograms. Also updated the way decorated input are created to automatically create the decorator inside the Set method. If the automatic computation of the minimum and maximum is enabled then streaming is disabled by the overloaded GetNumberOfInputRequestedRegions method.

The ImageSinc can be used as a replacement.

Use the double graft best practice for mini-pipelines in the NormalizeImageFilter.

blowekamp · 2019-05-16T14:37:09Z

This patch should be ready to go. I'm just refreshing the CI after one VNL test failure due to time profiling.

romangrothausmann · 2019-05-16T15:44:12Z

Trying to test this with my ITK-CLIs and am hitting the backwards incompatibility that the StatisticsImageFilter no longer produce an image as output. What would be the recommended way to adjust the code accordingly? Basically I would replace e.g.

stat->SetInput(extractor->GetOutput());
adder->SetInput1(stat->GetOutput());
adder->Update();
std::cerr << stat->GetMaximum();

by

stat->SetInput(extractor->GetOutput());
adder->SetInput1(extractor->GetOutput());
adder->Update();
stat->Update();
std::cerr << stat->GetMaximum();

Is there a more elegant way to trigger stat to update if adder is updated, similar to how it was possible before?

blowekamp · 2019-05-16T15:53:34Z

Great, thank you for testing!

I take it there is no streaming involved here. All you should have to do is simple call stat->Update(); seems elegant to me. 😎

romangrothausmann · 2019-05-16T15:57:28Z

Thanks for the feedback. No, no streaming there. I'll just use the additional stat->Update();
I am hitting this though with Module_ParabolicMorphology=ON:

/opt/itk/include/ITK-5.0/itkParabolicOpenCloseSafeBorderImageFilter.hxx:48:48: error: no matching function for call to 'itk::StatisticsImageFilter<itk::Image<unsigned char, 3u> >::GetOutput()'
     typename TInputImage::SpacingType spcing = m_StatsFilt->GetOutput()->GetSpacing();

That originates from:
https://github.com/InsightSoftwareConsortium/ITKParabolicMorphology/blob/d4bc69362c3e435ff5d97ed5bee0167199da69ac/include/itkParabolicOpenCloseSafeBorderImageFilter.hxx#L48

blowekamp · 2019-05-16T16:23:22Z

Here is a PR to improve that filter: InsightSoftwareConsortium/ITKParabolicMorphology#26

to be compatible with InsightSoftwareConsortium#855

romangrothausmann · 2019-05-17T15:32:40Z

@blowekamp I'm getting warnings like:

WARNING: In /opt/itk/include/ITK-5.0/itkImageSink.hxx, line 88
ImageToHistogramFilter (0x2955840): Unable to convert input "HistogramBinMaximum" to type N3itk5ImageIfLj3EEE

when using ImageToHistogramFilter of this PR with my hist see e.g. the log here for SDI:
https://gitlab.com/romangrothausmann/ITK-CLIs/-/jobs/214553760
and also if not using SDI:
https://gitlab.com/romangrothausmann/ITK-CLIs/-/jobs/214553759
It is unclear to my why and what to do about them. The result is still the same though.

hjmjohnson · 2019-05-21T20:26:04Z

@blowekamp I am also getting new warnings for every run of my application:

WARNING: In /Shared/johnsonhj/Binaries/20190513/Linux/NEP-14/ITKv5/Modules/Core/Common/include/itkImageSink.hxx, line 88
LabelStatisticsImageFilter (0x278c2d0): Unable to convert input "LabelInput" to type N3itk5ImageIdLj3EEE

WARNING: In /Shared/johnsonhj/Binaries/20190513/Linux/NEP-14/ITKv5/Modules/Core/Common/include/itkImageSink.hxx, line 88
ImageToHistogramFilter (0x279bd70): Unable to convert input "AutoMinimumMaximum" to type N3itk5ImageIsLj3EEE

WARNING: In /Shared/johnsonhj/Binaries/20190513/Linux/NEP-14/ITKv5/Modules/Core/Common/include/itkImageSink.hxx, line 88
ImageToHistogramFilter (0x279bd70): Unable to convert input "HistogramSize" to type N3itk5ImageIsLj3EEE

WARNING: In /Shared/johnsonhj/Binaries/20190513/Linux/NEP-14/ITKv5/Modules/Core/Common/include/itkImageSink.hxx, line 88
ImageToHistogramFilter (0x279bd70): Unable to convert input "MarginalScale" to type N3itk5ImageIsLj3EEE

`StatisticsImageFilter` is no longer an `ImageToImageFilter`, so it no longer supports `GetOutput()`, from pull request InsightSoftwareConsortium/ITK#855 InsightSoftwareConsortium/ITK@0efe8db "ENH: Add streaming support to StatisticsImageFilter", May 15, 2019

…reamingBaseClasses Add streaming base classes and streaming statistics

to be compatible with InsightSoftwareConsortium#855

blowekamp changed the title ~~WIP: Add streaming base classes~~ WIP: Add streaming base classes and streaming statistics May 10, 2019

blowekamp requested review from N-Dekker, dzenanz, hjmjohnson and thewtex May 10, 2019 19:46

dzenanz approved these changes May 13, 2019

View reviewed changes

blowekamp force-pushed the AddStreamingBaseClasses branch 5 times, most recently from 85e2edd to d12d5c0 Compare May 14, 2019 19:24

blowekamp commented May 14, 2019

View reviewed changes

Comment thread Documentation/ITK5MigrationGuide.md Outdated

blowekamp commented May 14, 2019

View reviewed changes

Comment thread Modules/Core/Common/include/itkImageSink.hxx

blowekamp requested a review from zivy May 14, 2019 19:33

dzenanz reviewed May 14, 2019

View reviewed changes

Comment thread Modules/Filtering/ImageStatistics/test/itkStatisticsImageFilterTest.cxx Outdated

Comment thread Modules/Filtering/ImageStatistics/include/itkStatisticsImageFilter.h

Comment thread Modules/Numerics/Statistics/include/itkImageToHistogramFilter.h Outdated

zivy reviewed May 14, 2019

View reviewed changes

zivy reviewed May 15, 2019

View reviewed changes

blowekamp force-pushed the AddStreamingBaseClasses branch from d12d5c0 to 87f4327 Compare May 15, 2019 16:12

blowekamp added 2 commits May 15, 2019 13:56

ENH: Add base class for generic streaming processes

f878f46

The StreamingProcesObject provide a pipeline mechanics to update the input's pipeline with a sequence of regions and generically execute them in a StreamedGenerateData method.

ENH: Add ImageSink base filter class

ac08bc1

Adding the ImageSink base class which provides support for streaming and multi-threading on an input image. Generally this is used when transforming an image into something like computing statistics, or a RLE of an image.

blowekamp force-pushed the AddStreamingBaseClasses branch from 87f4327 to d44a529 Compare May 15, 2019 18:29

blowekamp changed the title ~~WIP: Add streaming base classes and streaming statistics~~ Add streaming base classes and streaming statistics May 15, 2019

blowekamp commented May 15, 2019

View reviewed changes

Comment thread Testing/Data/Baseline/Filtering/ResampleImageFilter9TestPixelCentered.4.png.sha512 Outdated

blowekamp added 3 commits May 15, 2019 14:34

ENH: Use named input in LabelStatisticsImageFilter

29aa8e1

ENH: Add streaming support to StatisticsImageFilter

0efe8db

blowekamp added 7 commits May 15, 2019 14:34

ENH: Add ImageSink::VerifyInputInformation

c3fcfe2

And the same change from ImageToImgeFilter to verify that input images are congruent in physical space by default.

ENH: Add KWStyle override for ImageStatistics module

8f66190

ENH: Deprecate ImageTransformer class

08fcda0

The ImageSinc can be used as a replacement.

BUG: Update NormalizeImageFilter with double graft mini-pipeline

6c575b3

Use the double graft best practice for mini-pipelines in the NormalizeImageFilter.

DOC: statistics filter changes for ITKv5

efdca3a

ENH: Add Python wrapping for streaming base classes

a931acf

blowekamp force-pushed the AddStreamingBaseClasses branch from d44a529 to a17bb6e Compare May 15, 2019 18:35

ENH: Add streaming testing of label statistics filter

23da91e

blowekamp force-pushed the AddStreamingBaseClasses branch from a17bb6e to 23da91e Compare May 16, 2019 14:12

dzenanz approved these changes May 16, 2019

View reviewed changes

blowekamp merged commit c76b193 into InsightSoftwareConsortium:master May 16, 2019

romangrothausmann added a commit to romangrothausmann/ITK that referenced this pull request May 17, 2019

ENH: Bump remote module ParabolicMorphology

00516d6

to be compatible with InsightSoftwareConsortium#855

romangrothausmann mentioned this pull request May 17, 2019

ENH: Bump remote module ParabolicMorphology #891

Merged

romangrothausmann added a commit to romangrothausmann/ITK that referenced this pull request May 17, 2019

ENH: Bump remote module ParabolicMorphology

1e45056

to be compatible with InsightSoftwareConsortium#855

romangrothausmann added a commit to romangrothausmann/ITK that referenced this pull request May 17, 2019

ENH: Bump remote module ParabolicMorphology

2ce6cc1

to be compatible with InsightSoftwareConsortium#855

hjmjohnson mentioned this pull request May 21, 2019

Runtime warnings after #855 #938

Closed

This was referenced Jul 3, 2019

Update code to support ITKv5 multi-threading model blowekamp/itkStreamingSinc#4

Closed

using itkStreamingSinc in combination with itkImageToHistogramFilter blowekamp/itkStreamingSinc#2

Closed

hjmjohnson pushed a commit to hjmjohnson/ITK that referenced this pull request May 6, 2026

Merge pull request InsightSoftwareConsortium#855 from blowekamp/AddSt…

e21af0a

…reamingBaseClasses Add streaming base classes and streaming statistics

hjmjohnson pushed a commit to hjmjohnson/ITK that referenced this pull request May 6, 2026

ENH: Bump remote module ParabolicMorphology

2bdbc84

to be compatible with InsightSoftwareConsortium#855

Uh oh!

Conversation

blowekamp commented May 10, 2019

Uh oh!

blowekamp commented May 13, 2019

Uh oh!

dzenanz left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

dzenanz commented May 14, 2019

Uh oh!

blowekamp commented May 14, 2019

Uh oh!

zivy left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

romangrothausmann commented May 15, 2019

Uh oh!

zivy left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

blowekamp commented May 16, 2019

Uh oh!

romangrothausmann commented May 16, 2019

Uh oh!

blowekamp commented May 16, 2019

Uh oh!

romangrothausmann commented May 16, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

blowekamp commented May 16, 2019

Uh oh!

romangrothausmann commented May 17, 2019

Uh oh!

hjmjohnson commented May 21, 2019

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

romangrothausmann commented May 16, 2019 •

edited

Loading