Wrap LabelSetMeasures in LabelOverlapMeasuresImageFilter

[dzenanz]

This exposes the return value from itk::LabelOverlapMeasuresImageFilter::GetLabelSetMeasures() as a Python dictionary instead of <Swig Object of type 'std::unordered_map< unsigned char,itkLabelOverlapMeasuresImageFilterIUC3::LabelSetMeasures > *'>. This is needed for convenient use of the class from Python.

PR Checklist

• No API changes were made (or the changes have been approved)
• No major design changes were made (or the changes have been approved)
• Added Python wrapping to new files (if any) as described in ITK Software Guide Section 9.5

[dzenanz]

Currently running into:

3024: Loading ITKImageStatistics... done
3024: Running itkLabelOverlapMeasuresImageFilterIUC3... Running itkLabelOverlapMeasuresImageFilterIUC3... done
3024: done
3024: Traceback (most recent call last):
3024:   File "C:\Dev\ITK-git\Modules\Filtering\ImageStatistics\wrapping\test\itkLabelOverlapMeasuresImageFilterTest.py", line 33, in <module>
3024:     for label, measure in lsm.items():
3024:                           ^^^^^^^^^
3024: AttributeError: 'SwigPyObject' object has no attribute 'items'
3024: swig/python detected a memory leak of type 'std::unordered_map< unsigned char,itkLabelOverlapLabelSetMeasures,std::hash< unsigned char >,std::equal_to< unsigned char >,std::allocator< std::pair< unsigned char const,itkLabelOverlapLabelSetMeasures > > > *', no destructor found.
3024: itkTestDriver: Process exited with return value: 1

[thewtex]

What is dir(lsm)?

[dzenanz]

dir(lsm) is ['__class__', '__delattr__', '__dir__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__getstate__', '__gt__', '__hash__', '__init__', '__init_subclass__', '__int__', '__le__', '__lt__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', 'acquire', 'append', 'disown', 'next', 'own']

[thewtex]

@dzenanz could you please rebase and push? I am wondering if recent SWIG updates have helped.

[hjmjohnson]

Rebased onto current upstream/main (was 3894 commits behind). Both commits cherry-picked cleanly with only auto-merges in itkLabelOverlapMeasuresImageFilter.h and Wrapping/Generators/Python/PyBase/pyBase.i — no manual conflict resolution needed. CI will re-run on the new HEAD; the SWIG-update question @thewtex raised in May 2024 should now answer itself.

[hjmjohnson]

Pushed two commits on top of lsmLabelOverlap that take a different angle on the Python-side test failure.

Diagnosis (verified locally, macOS Python wrapping build): SWIG silently drops %template std::unordered_map<X, itk::LabelOverlapLabelSetMeasures> in the Instantiations.i because the value type is %import-ed (not %include-d) into the consuming submodule. Count of hashmap in the SWIG-generated ITKImageStatisticsPython.cpp was 0 across every namespace-qualifier and ordering variant tried. Confirmed both inside Modules/Filtering/ImageStatistics/wrapping/ and after relocating the %template declarations to pyBase.i — the second attempt failed for layering reasons (pyBase.i is foundation and can't see leaf-module headers).

Pivot to a wrapping-friendly C++ surface: added two small accessors that don't rely on std::unordered_map getting a Python-dict wrapper. The existing MapType GetLabelSetMeasures() is preserved unchanged for C++ consumers.

std::vector<LabelType>      GetLabels() const;
LabelOverlapLabelSetMeasures GetMeasureForLabel(LabelType) const;

Plus six Get* getters on the struct itself (igenerator doesn't auto-wrap public data members). Both work natively across the per-submodule SWIG isolation because std::vector<unsigned char> is already wrapped globally in pyBase.i and the struct is wrapped via the existing itk_wrap_simple_class directive.

Test now passes locally against the existing fixture:

Found 14 labels
Label: 1, i: 2246, u: 120691
Label: 2, i: 497,  u: 22835
…
Label: 13, i: 0,   u: 387126
1/1 Test #211: LabelOverlapMeasuresImageFilterTest ...   Passed    1.75 sec
100% tests passed, 0 tests failed out of 1

Commits (2, on top of your existing 2)

	Commit	Files
3	ENH: Add Python-friendly LabelSetMeasures accessors	itkLabelOverlapMeasuresImageFilter.h, itkLabelOverlapLabelSetMeasures.h
4	COMP: Drop unused std::unordered_map SWIG infrastructure	wrapping/CMakeLists.txt (drop WRAPPER_SWIG_LIBRARY_FILES), delete itkLabelOverlapLabelSetMeasuresInstantiations.i, revert pyBase.i %include <std_unordered_map.i>, update Python test

Both commits carry Co-Authored-By: dzenanz to preserve your authorship.

What I tried before pivoting (so this isn't a black box)

Variant	Result
Original (itk::LabelOverlapLabelSetMeasures qualified, in Instantiations.i)	%template silently dropped — 0 hashmap classes generated
Drop the itk:: namespace qualifier	SWIG generates templates but C++ can't compile (the flat name only exists as a typedef in the struct's submodule)
Add using itkLabelOverlapLabelSetMeasures = itk::LabelOverlapLabelSetMeasures; to Instantiations.i's %{…%}	Templates still dropped
Move %template to pyBase.i with %import "itkLabelOverlapLabelSetMeasures.h"	SWIG fails: header not on PyBase's include path
Forward-declare in pyBase.i (namespace itk { struct LabelOverlapLabelSetMeasures; })	SWIG generates code; C++ compile fails for layering reasons

Root cause is architectural: ITK's wrapping pipeline gives each submodule its own SWIG run with %import-only type info from siblings, and SWIG's %template std::unordered_map<X, Y> requires Y to be %include-d (full type definition) — not just declared. Fixing this cleanly would be a multi-day igenerator.py change with broad blast radius.

CI will re-run on 2e882122e2. If anything in this approach doesn't sit right with you, happy to iterate — the commits are surgical and easy to drop / amend.

[hjmjohnson]

Re-triggering ARMBUILD-x86_64-rosetta — itkImageAlgorithmCopyTest2 (3.14 float CheckBuffer) failed in isolation on this runner; same test passed on #6185/#6188/#6189 against the same runner pool, and CDash shows 0 test failures (Linux 3486/3486, Darwin 309/309, Linux.Python 176/176). Known x86_64-rosetta float-precision flake, unrelated to this PR's diff.

[greptile-apps]

Greptile Summary

This PR extracts LabelOverlapMeasuresImageFilter::LabelSetMeasures into a standalone LabelOverlapLabelSetMeasures struct with explicit Get* accessors, and adds GetLabels() + GetMeasureForLabel() methods to the filter, together with a new Python wrapping and test. This allows Python consumers to iterate label measures without hitting SWIG's inability to wrap std::unordered_map across submodule boundaries. The backward-compatibility typedef is retained under #ifndef ITK_FUTURE_LEGACY_REMOVE.

Confidence Score: 4/5

Safe to merge; only P2 style findings (misleading comment, extra blank line, no value assertions in test).

All findings are P2 — no logic errors, no API breaks, and the backward-compat alias is correctly guarded. Score reflects clean design with minor polish items.

itkLabelOverlapMeasuresImageFilter.h — misleading doc comment on the legacy typedef and an extra blank line.

Important Files Changed

Filename	Overview
Modules/Filtering/ImageStatistics/include/itkLabelOverlapLabelSetMeasures.h	New standalone header extracting LabelOverlapLabelSetMeasures struct with explicit Get* accessors for Python wrapping; well-structured and self-contained.
Modules/Filtering/ImageStatistics/include/itkLabelOverlapMeasuresImageFilter.h	Adds GetLabels() + GetMeasureForLabel() accessors for Python, moves struct to standalone header; copy-pasted misleading doc comment on legacy typedef and a double blank line.
Modules/Filtering/ImageStatistics/wrapping/itkLabelOverlapLabelSetMeasures.wrap	Minimal one-liner wrap file using itk_wrap_simple_class; correct approach for a plain struct.
Modules/Filtering/ImageStatistics/wrapping/CMakeLists.txt	Sets WRAPPER_SUBMODULE_ORDER to ensure LabelOverlapLabelSetMeasures is wrapped before the filter that uses it; correct ordering.
Modules/Filtering/ImageStatistics/wrapping/test/itkLabelOverlapMeasuresImageFilterTest.py	Python smoke test exercising GetLabels() + GetMeasureForLabel() and Get* accessors; no value assertions so numerical regressions would go undetected.
Modules/Filtering/ImageStatistics/test/Input/DzZ_Seeds.seg.nrrd.sha512	Valid 128-character SHA-512 hash for new test input file.
Modules/Filtering/ImageStatistics/test/Input/DzZ_T1.seg.nrrd.sha512	Valid 128-character SHA-512 hash for new test input file.
Modules/Filtering/ImageStatistics/wrapping/test/CMakeLists.txt	New wrapping test CMakeLists with proper guards for 3D uchar wrapping; correct structure.

Sequence Diagram

sequenceDiagram
    participant Py as Python caller
    participant F as LabelOverlapMeasuresImageFilter
    participant M as std::unordered_map (m_LabelSetMeasures)
    participant S as LabelOverlapLabelSetMeasures

    Py->>F: UpdateLargestPossibleRegion()
    F->>M: populate per-label entries
    Py->>F: GetLabels()
    F->>M: iterate keys
    F-->>Py: std::vector<LabelType>
    loop for each label
        Py->>F: GetMeasureForLabel(label)
        F->>M: find(label)
        M-->>F: iterator → LabelOverlapLabelSetMeasures
        F-->>Py: LabelOverlapLabelSetMeasures (by value)
        Py->>S: GetIntersection() / GetUnion() / …
        S-->>Py: SizeValueType
    end

Loading

_{Reviews (1): Last reviewed commit: "COMP: Drop unused std::unordered_map SWI..." | Re-trigger Greptile}

[dzenanz]

Looks good. I can't approve my own PR.