Page MenuHomePhabricator

Extend checklists for segmentation view
Open, HighPublic

Description

All tools have to check for the following usage scenario:

  • segmentation on a 3D image
  • 3D segmentation (static) on a 4D (dynamic) image
  • 4D segmentation on a 4D image
  • All this with an image which time geometry does not start at 0 ms (e.g. MITK-Data/3D+t-ITKIO-TestData/LinearModel_4D_arbitrary_time_geometry.nrrd)

What the segmentation should look like/how it is generated must be specified per tool. But for all tools the following post conditions should apply:

  • Tools can be activated, used /confirmed.
  • The timestep of the reference image stays the same and is not changed by starting the segmentation.
  • Using the tool does not change the geometry of the segmentation (so 3D stays 3D and 4D segs are still 4D segs after the tool usage).

Extend the segementation check lists accordingly.

Event Timeline

floca updated the task description. (Show Details)
floca added a project: Restricted Project.
floca moved this task from Backlog to Segmentation on the MITK (2020) board.
kislinsk renamed this task from Extend checklists for statistic view to Extend checklists for segmentation view.Jul 1 2020, 10:22 AM
kislinsk moved this task from Segmentation to Statistics on the MITK (2020) board.
kislinsk moved this task from Statistics to Segmentation on the MITK (2020) board.

I started with the checklists I from E130-Daten\Release\Checklisten\MITK Workbench Release.

Edit: I see this as a meta-task so I didn't claim this one. Please go ahead and choose / create a subtask for other checklists.

All tools have to check for the following usage scenario:

  • All this with an image which time geometry does not start at 0 ms (e.g. MITK-Data/3D+t-ITKIO-TestData/LinearModel_4D_arbitrary_time_geometry.nrrd)

For this I would like to refer to T26989: [Checklist] GUI test data originates from different sources to discuss a new set of test images.

Damn I have forgotten to copy the checklist to a reachable share. @kalali Is it already done now or obsolete?

The checklist don't say anything about multilabel segmentation. This should be addressed with a special focus on creating and combining multiple labels / layers.
I couldn't find any other checklist with a focus on the multilabel functionality.

OK. But I tend/think more and more that we should strive for a unit test testbed that tests all that on the tool class level (either the respective base classes or the tools itself where it is easily possible (e.g. because the have no interaction (but even thous shouldn't be a show stopper).

So before anyone does extensive checklist rewriting or checking, please evaluate the possibility to directly invest more into unit tests.

I'm all in for that. We already discussed that we will focus again on this task T26754: Review checklists and its subtasks. This also includes

Automatic GUI testing

and

move as much tests as possible to Unit-Tests

So I'll agree. After the existing segmentation checklists have now been re-checked, we could use them and start reducing the manual tests there, as mentioned

So to reduce the size and complexity of the checklist I think we should try to

  1. remove tests that can be automated
  2. group / separate tests according to a reasonable similarity / dissimilarity
  3. split the document into the clustered algorithms / continue with {T26987} for a better checklist format