Using --no_norm with CLI slows down 3D segmentation[BUG]

Open derekthirstrup opened this issue 1 year ago • 1 comments

The segmentation of 3d tif stacks is taking 5-6 minutes to process 1 image when --no_norm is used. When the same stacks are processed without --no_norm argument then it takes roughly 20 seconds. I previously denoised and rescaled the images so they do not need to be rescaled again during segmentation. Any ideas why it takes significantly more time with --no_norm option? Seems like it should be faster.

Mar 22 '24 20:03 derekthirstrup

hmm this is strange, is the output exactly the same with and without --no_norm?

Sep 11 '24 07:09 carsen-stringer

this could be due to type-casting, if your images are not float32, we've updated the type-casting to occur on the GPU which may speed things up, but let us know if this is still an issue

Oct 26 '24 13:10 carsen-stringer