Open m-albert opened 2 months ago
I've also experienced the same problem recently.
No advice yet, I haven't figured out what is different this time compared to the last time I tried to use this (quite some time ago)
This issue https://github.com/GrotjahnLab/surface_morphometrics/issues/24 from two weeks ago has the suggestion to try:
Thanks Genevieve!
I had missed that issue and will try the two suggestions and report back.
Okay, so even with a lot of available memory (128GB) and max_triangles=50000
(a third of the default value) I'm getting the segmentation faults. Also SLURM reported that only ~300MB were used by the failed jobs.
The segmentation faults appear just after the curvature warnings:
2024-07-01 14:50:19.936 ( 4.617s) [ 7F03A9204740] vtkCurvatures.cxx:379 WARN| vtkCurvatures (0x55ebe9fac7f0): The Gaussian or mean curvature at point 26255 have a large computation error... The maximum curvature is likely off.
2024-07-01 14:50:19.936 ( 4.617s) [ 7F03A9204740] vtkCurvatures.cxx:379 WARN| vtkCurvatures (0x55ebe9fac7f0): The Gaussian or mean curvature at point 26256 have a large computation error... The maximum curvature is likely off.
2024-07-01 14:50:19.936 ( 4.617s) [ 7F03A9204740] vtkCurvatures.cxx:379 WARN| vtkCurvatures (0x55ebe9fac7f0): The Gaussian or mean curvature at point 26257 have a large computation error... The maximum curvature is likely off.
2024-07-01 14:50:19.936 ( 4.617s) [ 7F03A9204740] vtkCurvatures.cxx:379 WARN| vtkCurvatures (0x55ebe9fac7f0): The Gaussian or mean curvature at point 26258 have a large computation error... The maximum curvature is likely off.
2024-07-01 14:50:19.936 ( 4.617s) [ 7F03A9204740] vtkCurvatures.cxx:379 WARN| vtkCurvatures (0x55ebe9fac7f0): The Gaussian or mean curvature at point 26259 have a large computation error... The maximum curvature is likely off.
srun: error: maestro-1016: task 0: Segmentation fault
I've experienced segmentation faults before when using vtk
in combination with meshes, but I don't remember to have found a solution or pattern to when the problem occurred. I wouldn't be surprised if it correlated with meshes sometimes having weird properties like intersecting faces and such...
So it looks like the input surfaces I was making are malformed vtk files some of the time. Something is wrong with this new meshing protocol - I am working on it. In the mean time, try turning "ultrafine" off in the config. I'll hopefully have a fix in the next couple days - we are now reproducibly getting this issue in-house which of course makes it much easier for me to fix!
Thanks a lot for looking into this @bbarad! Happy to help with any testing from 15/7 on.
Thanks for working on a fix! I am also very happy to help test things.
I've narrowed things down to a new issue with either pymeshlab's ply writer, vtk's ply reader, or vtk's vtp writer (seems least likely!). The VTP file that gets fed into pycurv etc is malformed - paraview can't load it either. Would either of you mind sharing a pip freeze of your environment that is causing segfaults?
OK - I've narrowed it down to just VTK operations, as the ply file that is generated seems to work perfectly with other analysis.
Disregard that - any VTK that I make from one of these ply files won't work. Seems to be coming from the pymeshlab side...
Would either of you mind sharing a pip freeze of your environment that is causing segfaults?
Sadly, setting "ultrafine: false" did not help.
Yeah, I can't get anything from meshlab or pymeshlab through with any settings on the workstation. Works fine on my laptop of course... Just a matter of identifying the difference now!
On Mon, Jul 8, 2024 at 6:03 PM Genevieve Buckley @.***> wrote:
Sadly, setting "ultrafine: false" did not help.
— Reply to this email directly, view it on GitHub https://github.com/GrotjahnLab/surface_morphometrics/issues/27#issuecomment-2215919365, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABKJDT4XU54DTRRNMXLZEJDZLMZFTAVCNFSM6AAAAABJ7Q2VWKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMJVHEYTSMZWGU . You are receiving this because you were mentioned.Message ID: @.***>
You said you have a reproducible example now, would it help to share that so we can both be looking at the same thing?
Although I not running on a HPC. Some of the tomograms succesfully run the run_pycurv.py step and some don't. So I will also share my pip freeze and conda list
I am using the surface_morphometrics-1.0beta1 version
I am actually running into my problems with the example data, which shows its really generic. Its happening with or without the new ultrafine sampling, so I don't believe thats the issue. I've attached the ply and surface.vtp files. What's odd is that both of these seem to work fine on my m1 mac but not on my ubuntu workstations... Surfaces.zip
The issue always occurs to me when I am running it on the Inner mitochondria membrane and always works with the OMM without any issues!
But wow, were you able to find a workaround or a solution that could work?
Hi,
I'm occasionally getting segmentation faults while running
python run_pycurv.py config.yml ${i}.surface.vtp
as part of the configurable pipeline. However, the meshes created duringpython segmentation_to_meshes.py config.yml
look good judging by visual inspection.Before going into debugging, I was wondering whether this is would perhaps be a known problem and whether there'd be some go-to parameters / things to look out for for avoiding the seg faults?
Thanks!