To reproduce:
-> workbensh crashes in QmitkMultiLabelSegmentationView::OnLooseLabelSetConnection() at assert(workingImage);
| rMITK MITK | |||
| Restricted Differential Revision | rMITKb547a8f6352b Fix newly introduced memory leak | ||
| Restricted Differential Revision | rMITK502bcaa86770 Fix newly introduced memory leak | ||
| Restricted Differential Revision | rMITK415b55807ca4 Add auto-crop feature for LabelSetImages | ||
Checked it the source of the problem is QmitkAutocropAction of the multisegmentation plugin. It generates an mitk::Image instead of a mitk::LabelSetImage as a result. As a consequence a precondition of the QmitkMultiLabelSegmentationView fails with an assertion.
The action was never implemented correctly :(.
I would propose for now to completly deprecate and remove this action. It would a major rework and I also think it is without knowing the concrete use case also illposed as context menu action, as there are many implecations. Do you want to crop just a specific label (which one?)? Crop only a laxer (wich one?)? Crop the entier label set (often not very meaningfull...) I think such a tool is of use, but should be (just) part of the Segmentations Tools View. But in context of this task, I would just remove the action. There implementation based on Toolproviders is anyway shacky/shady as discussed in T27046.
Remark: There is some code in QmitkLabelSetWidget that implements the creation of a cropped mask for one lable (at least).
Discussion result: Stefan will have a look if he might find an easy solution ;) Otherwise skipp it.
@kislinsk What's the status? We discussed this and the mentioned task last Friday and decided to disable the autocrop-freature for multi-label via data manager context menu if there is no easy solution.
There are two separate issues:
The second issue is easy to fix as it boils down to a missing nullptr check. If the signal for a lost working node was triggered, it is possible that the data of that node isn't already a LabelSetImage anymore, so we simply have to check the type. It's not the scope of this task but I question the general logic of QmitkMultiLabelSegmentationView::OnLooseLabelSetConnection(). It basically removes observers from the working node and it does not crash for LabelSetImages but I guess the possibility is high that it is already another LabelSetImage and we don't want to remove observers from the new LabelSetImage. Maybe it is a working-by-accident thingy and observers are added shortly after again.