constantinpape/torch-em

Issues with PlantSeg dataset

Closed this issue · 1 comments

anwai98 commented

I have observed the following issues (under the Root subset):

  • Some test volumes have a few empty patches towards the end of the stack
  • Some volumes seem to be repeated across splits (eg. Movie1_t00045_crop_gt.h5). It could surely be a splitting / naming issue, need to double check this.

PS. Also double-check the aforementioned issues in other data subset in PlantSeg.

anwai98 commented

Okay, I verified the above points, and:

  • Re: empty patches or artifacts in some volumes - seems like it's the case for quite a few volumes in root, and some in ovules as well.
  • Re: repeated volumes in train and test for root - it seems like both the volumes are same. I went ahead and removed it as well.

I went ahead and fixed both issues in the aforementioned PR.