-
-
Notifications
You must be signed in to change notification settings - Fork 2.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Output of rescale
different from 0.18.3 to 0.19.0
#6093
Comments
I assume you also have SciPy >= 1.6.0? If so, then the following is likely the source of this change: scikit-image/skimage/_shared/utils.py Lines 699 to 705 in 75f532b
Note, that the user-specified "constant" gets converted to SciPy's "grid-constant" and "wrap" gets converted to "grid-wrap". In the case of wrap, the SciPy code is not what one would expect, so I would consider "grid-wrap" a bug fix rather than unintentional change. For "constant" vs. "grid-constant" it is more a matter of convention than a true bug fix so we can debate if this change should have been made. The difference between the two is shown in the top two rows of this figure from the SciPy docs: |
This decision introduced a silent breaking change with no easy fix for our users (please see sunpy/sunpy#5750). Maybe should we rollback this decision and simply propose |
Yeah, that seems fine. |
Actually, on closer look, I am not sure we should change this on the v0.19.x branch. If I take the SciPy 1.6-based code path vs. commenting that out so that the alternative, legacy 2D code path is taken, the images are the same (to within ~1e-11 floating point precision). So the new That said, I could reproduce the difference from 0.18.3 to 0.19.0 as reported here, although I am not yet sure of the cause. The "legacy" code path linked above looks the same as the v0.18.3 code. |
Okay, I understand now. Our own cython-based interpolation code is only called within However, we should not change given that, I will close this issue as I consider the current change to be a bug fix |
Description
The output of
skimage.transform.rescale
has changed from version 0.18.3 to 0.19. The changes all appear to be around the edge of the output - see below. As far as I can tell this change isn't documented anywhere in the release notes (https://scikit-image.org/docs/stable/release_notes.html), so I was wondering if this was either a bug, an undocumented deliberate change, or if I've missed something in the release notes?Way to reproduce
Version information
The text was updated successfully, but these errors were encountered: