Most elements in default data is 0, so a large number of inputs is full 0 assay.

[paperplane03]

Dear Staff,
I participated the contest and run the 'python train_3D.py' in README. I followed the instruction (fetch the 36.7G data and get an environment) and find most of the inputs is 0. Is it normal of something gets wrong?
I calculated the fraction of 0 in 'inputs' by coding in './cellmap-segmentation-challenge/src/cellmap_segmentation_challenge/train.py' and find most patches's fraction is 1.

[rhoadesScholar]

Hi @paperplane03. Yes, most arrays are all 0 (or False). This is because almost all crops do not have all organelles present. :)

[paperplane03]

Thank you!

[Luchixiang]

Is there any way to only crop the image region with the ground truth labels and ignore those are zeors?

[paperplane03]

So do I. Is there any ways to crop only non-zero input?

[rhoadesScholar]

Is there any way to only crop the image region with the ground truth labels and ignore those are zeors?

If you use the included data loading utilities, only regions with data should be loaded. Otherwise, you can consider using the training_crop_manifest to export cropped versions of the raw data.