You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
scikit-image 0.26 has deprecated binary_erosion, binary_dilation, binary_opening and binary_closing. The greyscale morphology functions can cover the binary case. So far we did not do the same deprecation in cuCIM.
For cuCIM, I think the kernel performance is a bit better in the binary case, so just using the greyscale kernels (based on minimum/maximum filtering) is slower. So if we just naively do the deprecation it will cause a performance regression. We can still consider the deprecation, but in order to do it, we should keep the underlying kernel implementation and automatically dispatch to the faster binary kernel when possible from the greyscale morphology functions.
cupyx.scipy.ndimage also provides binary morphology functions, but they do not have built-in support for passing in a sequence of structuring elements (from cuCIM/skimage structuring element decomposition).
Activity