Gregory Lee

Thanks for bringing this up @rjesud. `regionprops` is one area that is currently known to not be accelerated well in cuCIM. What are the typical "areas" of the regions in...

@gigony, should we add a `performance` label? I wouldn't classify this as a bug in that the output is correct, it is just slower than desired.

In the specific case of "area", since this is just the number of pixels, I think you can use `cupy.bincount` to more efficiently get all of the areas in one...

Here is a function that should work specifically for areas and centroids. It assumes the background has label value 0 and all other integer values are labeled regions. It is...

> The other feature I am interested in is mean_intensity. Do you have any suggestion for a work around? Can you try [`cupyx.scipy.ndimage.mean`](https://docs.cupy.dev/en/stable/reference/generated/cupyx.scipy.ndimage.mean.html#cupyx-scipy-ndimage-mean) and see if it is sufficient? (or...

> In other runs, I see it take anywhere from 4 to 20 seconds. I am not sure what causes this difference. Is there some sort of cache I should...

Hi @FrancescoRossiAiko, adding this along with other functions under `skimage.filters.rank` is something I want to do. These would share a lot of common code and I think this would be...

> Do you think there may be a way to collaborate with or to help you? Sure, we are happy to have contributions. Do you have some familiarity with writing...

To clarify, the implementation would live here and not in CuPy itself. We would just need to vendor a copy of the kernel generation code at this location https://github.com/rapidsai/cucim/tree/branch-22.04/python/cucim/src/cucim/skimage/_vendored because...

@jakirkham, it is just that we have not implemented it here. There are actually two underlying Cythom implementations in scikit-image. The faster of the two uses integral images to make...