Repository navigation
Keyword *interpolation* behaving improperly while saving to SVG with **savefig()** #12065
Description
Bug report
Bug summary
When saving an interpolated rbga image to SVG format, the output depends whether the keyword interpolation as been omitted or it has been put to none (interpolation="none"). This should not be the case. It should produce an UN-interpolated image in both case. We do not see this behavior with the other formats, PNG for instance.
Code for reproduction
import matplotlib.pyplot as plt
import numpy as np
fig, (axl, axr) = plt.subplots(1, 2)
# full green image
img = np.zeros((5, 5, 4))
img[...,1] = np.ones((5, 5))
# transparent above main diagonal
img[...,3] = np.tril(np.ones((5, 5), dtype=np.uint8))
axl.set_title('with keyword\ninterpolation="none"')
axr.set_title('WITHOUT keyword\ninterpolation')
axl.imshow(img, interpolation="none")
axr.imshow(img)
plt.savefig('interp_alpha_svg.svg')
plt.savefig('interp_alpha_png.png')Actual outcome
Github does not support SVG so here is my description. The left image, with interpolation="none", looks interpolated (smoothed edges), but the right image, WITHOUT the interpolation keyword, has sharp edges.
Expected outcome
This is the output you get when format is PNG.
Matplotlib version
- Operating system: Ubuntu 16.04 Bash on Windows10
- Matplotlib version: 3.0.0rc1.post186+g5e01393
- Matplotlib backend (
print(matplotlib.get_backend())): agg - Python version: 3.5.2
- Jupyter version (if applicable):
- Other libraries:
matplotlib installed from source
Activity
anntzer commentedon Sep 7, 2018
I think this is basically an instance of #10112: the uniniterpolated image is stored as a big 5x5 image and svg renderers interpolate it by default; the non-interpolated image is upsampled (in nearest mode) to a much larger number of pixels.
dabana commentedon Sep 7, 2018
Okai so if I understand correctly, if we want to fix this issue we need to make sure that a large UPsampled image gets sent to the SVG renderer with interpolation="none" as well as WITHOUT the interpolation keyword (which is already the case).
anntzer commentedon Sep 7, 2018
No, we want to tell the svg renderer not to perform resampling itself, as explained in #10112.
dabana commentedon Sep 7, 2018
Just so I get this clear, specifying interpolation="none" and omitting interpolation behaving differently IS the intended behavior?
anntzer commentedon Sep 7, 2018
Good point, that I don't know...
dabana commentedon Sep 7, 2018
IF "specifying interpolation=none" and "omitting interpolation" behaving differently IS the intended behavior
THEN, oh well... close this issue...
OTHERWISE we have two options:
A) We should make sure that a NON-sampled image gets sent to the SVG renderer with interpolation="none" (which is already the case) as well as WITHOUT the interpolation keyword. That way the SVG render takes care of doing whatever it wants.
B) We should make sure that the SVG renderer always does a descent job by sending it an UP-sampled image with interpolation="none" as well as WITHOUT the interpolation keyword (which is already the case).
A or B is the question now...
dabana commentedon Sep 7, 2018
I vote for B
ImportanceOfBeingErnest commentedon Sep 7, 2018
Just that we know what the issue is, here is the output of the code when the svg is opened in inkscape 0.91:
(Maybe worth noting that inkscape 0.49 shows no difference between the two subplots)
ImportanceOfBeingErnest commentedon Sep 7, 2018
Equally worth noting that there is a difference between
Noneand"none"(Noneis the image.interpolation rc parameter, which defaults to"nearest").The interpolation_methods demo states:
So apparently some difference is expected between the two.
dabana commentedon Sep 8, 2018
From the imshow() documentation:
So it IS the intended behavior. In the doc, should we specify:
image.interpolationis the same as 'nearest'timhoffm commentedon Sep 9, 2018
The docstring has already changed a bit in the current master:
https://matplotlib.org/devdocs/api/_as_gen/matplotlib.axes.Axes.imshow.html
Issue 2 is already handled. 1 an 3 remain open. You are welcome to provide a PR to clarify the SVG handling.
clarifications to **im_show()** doc regarding issues matplotlib#12065 …
Update backend_svg.py