Gregory Lee
Gregory Lee
One caveat is potential considerations due to numerical stability. Using a two-pass approach is likely more accurate in general, so perhaps we should still provide an option to use that...
> It looks like tools/precompute/moments_sympy.py needs a little cleanup (print statements). This was intentional (see https://github.com/scikit-image/scikit-image/pull/6188#discussion_r788213676)
Rebased and updated the test cases that were added in the meantime for anisotropic spacing. As for other cases, we only want to compare the moments for the upper left...
@lagru, I added a 0.20 tag here in hopes we can consider this one for the release. The summary of benefits are: - performance improvement (a bit less than a...
> Perhaps we could https://github.com/cython/cython/issues/1772#issuecomment-371484338 in [used_numerics.pxd](https://github.com/scikit-image/scikit-image/blob/1ede87167068c9e8fc8871baf6a80fcce20eabe9/skimage/_shared/fused_numerics.pxd)? That would be a way to address this now while keeping track of the "proper fix" in TODO.txt. Yeah, this seems worth trying....
@mkcor, I fixed the conflict here and it looks like I had previously committed your renaming suggestion. Was there anything else remaining on this one?
Thanks for reporting this @efmkoene. We will take a look (please open a PR if you like!)
Thank you for looking into this @efmkoene! > I'm willing to make the changes (e.g., compute the appropriate Hessian using convolutions with Gaussians), but I am worried about making changes...
> For \sigma values
closed by #6446