BUG: fix crash in image normalization for RGBA image arrays #4094
Conversation
| if data.ndim == 3: | ||
| assert data.shape[-1] == 4 | ||
| # this is an RGBA array, only linear normalization makes sense here | ||
| norm_type = Normalize |
There was a problem hiding this comment.
If it's RGBA then I'm not sure I understand in what cases we'd want normalization. If it's not already scaled 0..1 or 0..255, then we need to apply that manually, and I don't think we want colorbars for that.
There was a problem hiding this comment.
indeed, this is just the simplest patch to restore previous behaviour. Before merging #3849 we used to pass a Normalize object in _init_image, and MPL didn't complain, because it's what it uses by default.
A different approach would be to patch this in _init_image, though I have a strong preference for leaving this responsability entirely to the NormHandler class because it's its sole purpose. Does it make sense ? Maybe I can add a little more details in the comment here ?
|
@yt-fido test this please |
neutrinoceros
force-pushed
the
hotfix_rgba_image_norm
branch
from
617ab0b to
7ec78b6
Compare
|
@yt-fido test this please |
neutrinoceros
force-pushed
the
hotfix_rgba_image_norm
branch
from
7ec78b6 to
db68135
Compare
|
@matthewturk if you get the chance, please review this again |
PR Summary
Fix #4093
the bleeding edge CI job is running on my fork at https://github.com/neutrinoceros/yt/actions/runs/2911689231