BUG: Preserve OutputRegionMode in IterativeDeconvolutionImageFilter

[physwkim]

Stop IterativeDeconvolutionImageFilter::GenerateData() from overwriting the output region metadata that GenerateOutputInformation() had already computed, which made OutputRegionMode a no-op. Addresses B17 of #6575.

Root cause

itkIterativeDeconvolutionImageFilter.hxx:113-116:

outputPtr->SetRequestedRegion(inputPtr->GetRequestedRegion());
outputPtr->SetBufferedRegion(inputPtr->GetBufferedRegion());
outputPtr->SetLargestPossibleRegion(inputPtr->GetLargestPossibleRegion());

ConvolutionImageFilterBase::GenerateOutputInformation() (itkConvolutionImageFilterBase.hxx:39-45) has already shrunk the output's largest-possible region for OutputRegionModeEnum::VALID by the time GenerateData() runs — and these three lines overwrite it with the input's (mode-blind) regions. PadInput and CropOutput (itkFFTConvolutionImageFilter.hxx:157, :468) read the output's requested region after the overwrite, so SetOutputRegionModeToValid() was accepted and silently ignored by Landweber, ProjectedLandweber, and RichardsonLucy alike.

The requested region is now derived from the output's own largest-possible region, which already reflects the selected mode. The manual buffered-region assignment and Allocate() were redundant: CropOutput() calls AllocateOutputs(), which sets the buffered region and allocates.

Local validation

New GoogleTest itkIterativeDeconvolutionImageFilterGTest.cxx.

Fail-before (OutputRegionModeValidShrinksOutputRegion): the filter returned the full 10x10 region instead of the expected 8x8 VALID-mode region.

Pass-after:

1/3 IterativeDeconvolutionImageFilterTest.OutputRegionModeSameMatchesInput ......... Passed
2/3 IterativeDeconvolutionImageFilterTest.OutputRegionModeValidShrinksOutputRegion .. Passed
3/3 ProjectedIterativeDeconvolutionImageFilterTest.BasicObjectMethods ............... Passed
100% tests passed

ctest -R Deconvolution (33 tests): identical results with the fix applied and reverted — the 14 legacy baseline tests fail identically in both, on ExternalData images this offline build tree never fetched (ExternalData_URL_TEMPLATES empty in the cache), i.e. before GenerateData() is reached. No baseline result changed because of this fix; CI will exercise them for real.

AI assistance

• Tool: Claude Code
• Role: implementation and regression-test authoring from the B17 analysis in BUG: Findings collected while porting ~60 filters — reported together in one issue #6575
• All code was reviewed, built, and tested locally before committing

[greptile-apps]

Greptile Summary

This PR preserves the deconvolution output region chosen by the convolution base class. The main changes are:

• Uses the output largest possible region when setting the requested region in GenerateData().
• Removes manual input-region copying and early output allocation from iterative deconvolution.
• Adds GTest coverage for SAME and VALID output-region modes.

Confidence Score: 5/5

This looks safe to merge.

No blocking issues found in the changed code.

No files need attention.

T-Rex Logs

What T-Rex did

• Saved the earlier raw tooling probe artifact for iterative_deconvolution to establish baseline context.
• Captured the main tooling blocker check results showing cmake_exit:127 and ninja_exit:127, and that the intended configure/build/ctest commands were not run.
• Verified that the C++ compiler version check completed successfully.

_{Ran code and verified through T-Rex}

Important Files Changed

Filename	Overview
Modules/Filtering/Deconvolution/include/itkIterativeDeconvolutionImageFilter.hxx	Updates GenerateData() to preserve output-region metadata computed before execution.
Modules/Filtering/Deconvolution/test/CMakeLists.txt	Adds the new iterative deconvolution GTest source to the module GTest driver.
Modules/Filtering/Deconvolution/test/itkIterativeDeconvolutionImageFilterGTest.cxx	Adds tests for SAME and VALID output-region behavior on Landweber deconvolution.

_{Reviews (1): Last reviewed commit: "BUG: Preserve OutputRegionMode in Iterat..." | Re-trigger Greptile}

[hjmjohnson]

A high-effort automated review of this branch raises design questions about the VALID-mode path that deserve author input before merge (force-push so far = plain rebase on main only).

Findings (ranked)

1. Zero-size VALID region now throws (CONFIRMED): with a kernel larger than the input (e.g. 3×3 image, 7×7 PSF), GetValidRegion() is size-0 and the new SetRequestedRegion(LPR) feeds a zero-size extraction into ExtractImageFilter, which throws — previously such pipelines produced (arguably meaningless) output without error. Worth an explicit itkExceptionMacro with a clear message, or documented rejection.
2. Pipeline convention: forcing the output requested region inside GenerateData() bypasses region negotiation; the pipeline-sanctioned hook for "this filter always produces its full output" is an EnlargeOutputRequestedRegion() override.
3. VALID/SAME numerical inconsistency (PLAUSIBLE): the requested-region change makes PadInput crop the kernel-radius padding before the FFT, and since iterative deconvolution is global in the FFT domain, VALID output is not the boundary-independent interior of the SAME output — users comparing the two get different values everywhere. If intended, worth documenting.
4. Stale estimate cache (PLAUSIBLE): Initialize() keys m_CurrentEstimate only on input MTime; switching OutputRegionMode after an aborted run reuses a padded estimate with the wrong geometry.
5. Test blindness: both new tests use a constant image + constant kernel and assert region metadata only, so a spatially shifted VALID crop would pass; only odd 3×3 kernels are exercised (the even-kernel and kernel-larger-than-image branches of GetValidRegion() are untested). A ramp input plus one even-kernel case would close this.

[hjmjohnson]

Force-push = plain rebase onto current main, folding the new GTest source into the GTests list that #6589 introduced for this module (two creategoogletestdriver calls in the CI merge build caused/would cause a duplicate-target configure error). Verified locally: configure clean, module tests pass.