Why ResampleImageFilter is slow?

(Bradley Lowekamp) #41

We should take this one issue at a time.

It would be very convenient to me if there was a pull request which demonstrated the issue, so that I could easily reproduce the problematic scenario.

(Matt McCormick) #42

Sometimes this error can be avoided by calling UpdateLargestPossibleRegion() instead of Update().

(grothausmann.roman) #43

Many thanks Matt and Brad for your replies. I created a PR (#82) but it seems to fail on some test I did not touch. I’m also not sure how to contribute the result MHAs/MD5s in conjunction with a GH PR. What are the instructions for that?
@blowekamp Would you prefer a separate PR for the test branch (test_RSI-SDI) without the contributions to itkResampleImageFilter?
@matt.mccormick I also tried streamer->UpdateLargestPossibleRegion() but that did not change the output of the monitor filter, i.e. make the test succeed. I can’t see why it is not working where as it does in itkWarpImageFilterTest2.cxx, which I used as an exemplary base.

(Matt McCormick) #44


It looks like the PR builds are failing because of the missing testing data.

Please try out this testing data script, and let us know how it goes:


(grothausmann.roman) #45

Many thanks @matt.mccormick for the script. It worked with a GIRDER_API_KEY (it was not able to authenticate with username and password). I added the *.sha512 and removed the *.md5 and sqaushed, rebased and force pushed all to GH.
Does it matter that the MHAs are not listed under ITK/ITKTestingData/4.13.1/Modules/Filtering/ImageGrid/test/Baseline on data.kitware.com?
Anyway, circleci now gets further and fails on some test one of which is itkResampleImageTest7 (as expected).
@blowekamp For better understanding the ITK GH workflow: In what way is it helpful for you to have a PR in this case? So far I was only considering to create PRs in the case all is basically ready for merging and tests work out on the local machine.

(Matt McCormick) #46

Thanks for the testing and feedback!


Nice work.

We will have the script remove likely invalid *.md5 content links automatically.

No, they can be anywhere, including your personal folder. This helps keep the red tape down. We will have an automated script upload merged testing into the ITK/ITKTestingData/Nightly folder, and at releases we will distribute the data to additional data storage locations.

(grothausmann.roman) #47

I created a separate PR for the branch with just the new tests (https://github.com/InsightSoftwareConsortium/ITK/pull/84), which nicely shows that there only itkResampleImageTest7 fails. So the other failures in https://github.com/InsightSoftwareConsortium/ITK/pull/82 are due to the changes to the itkResampleImageFilter.
So far I could not make out the reason for the change in the output extent which sometimes appears when streaming itkResampleImageFilter. Could there be some code that increases the demanded output extent when streaming?