2D Segmentation CNN Postprocessing Dimensions Error

[FliFlaFlock]

I am currently trying 2D segmentation and am running against a wall. The program takes a very long time at the CNN_postprocessing, namely at the refactoring to the original size (while otherwise being quick). I believe the problem is due to it trying to convert a 1x5172x5172 image to a 1024x1024 image as you can see in the attached output (script.txt). After waiting for a long time the program does continue but kills itself shortly after.

I am really unsure where the extra dimension comes from. At the pre-processing step it gains a dimension with value 1 (I am guessing due to all the other pars having a dim of 3), but that is later removed (from 1x1x5172x5172 to 1x5172x5172).

In regards to 2D segmentations I could only find that the z-component of the patch size should be 1, as I have applied. I also changed the first (z) value of the patch_halo to 0 as other values threw errors.

I attached the tif (single slice of a stack) as well as the final z-projection I would like to try it with along with the cmd line output and yaml. Both tifs produce the error.

I am using 1.8.1 on ubuntu 20.04, installed via conda, cuda enabled.

Best,
Constantin

PlantSeg_2D_Issue.zip

[qin-yu]

Thanks for reporting @FliFlaFlock

We are actively working on the release of PlantSeg v2. If this problem came out from --gui then this will no longer be supported; if this came out from --napari then it has very likely been fixed. Which interface did you use? I'll look into it.

[FliFlaFlock]

Thanks for the relply @qin-yu

I tested it for both cmd-line (using a config file) and the --gui. I did not test it using napari.

[qin-yu]

I came back to this issue again today. I believe the program was killed during saving not rescaling. @FliFlaFlock do you still encounter the same issue, or did you try v2.0.0aX?