Repository navigation
[Bug]: Irregular pixel widths in rasterized pcolor #23391
Description
Bug summary
Pixel widths appear irregularly in pcolor(x, y, z, rasterized=True), while x and y are regular. Using rasterized=False provides correct (equal size) pixels, but saving the figure to .png (instead of .pdf) will show the same irregular pixel width bug.
It looks like a resampling bug, which messes up precise data visualization completely.
Code for reproduction
import numpy as np
import matplotlib.pyplot as plt
x, y = np.meshgrid(range(500), range(2))
z = [np.tile([0,1],250), np.tile([0,1],250)]
fig, ax = plt.subplots(1,1, figsize=(10,2))
im = ax.pcolor(x, y, z, rasterized=True)
ax.set_xticks(np.arange(0,40))
fig.savefig('test.pdf')
plt.show()Actual outcome
In the resulting plot, some pixels are thinner, and some are thicker, than they should be.
rasterize=True (zoom-in to the lower left corner)
Expected outcome
rasterize=False (zoom-in to the lower left corner)
Additional information
- This behaviour depends on the number of pixels plotted, for me it becomes visible above
len(x)>100. - The spread between thicker and thinner lines increases with increasing size of the input array.
- If the view is reduced to a lower number of pixels (e.g.,
ax.set_xlim(0,10)), the effect is no longer visible.
Operating system
Windows
Matplotlib Version
3.5.1
Matplotlib Backend
inline
Python version
3.9.6
Jupyter version
No response
Installation
pip
Activity
jklymak commentedon Jul 6, 2022
I think this is basically #17182
jklymak commentedon Jul 7, 2022
I'm going to close as an understood problem tracked in #6827 among other places. Yes this is very annoying, and there is a partial solution that we hope to implement soon... Thanks,
mschroen commentedon Jul 7, 2022
Thanks for working on a solution. The referenced issues seem to deal with border pixels and overlapping, rather than constant size pixels, but I understand if this is related at the very bottom of the code implementation. Nevertheless, I don't see why this is a reason to close (~mark as solved) this issue?
jklymak commentedon Jul 7, 2022
We have over 1.5k issues so we try and cut down on duplicates at the triage stage.
Also for the behaviour you are describing, you should play with the various places "snap" can be set. https://matplotlib.org/stable/api/_as_gen/matplotlib.artist.Artist.set_snap.html
tacaswell commentedon Jul 7, 2022
Try passing
which should increase the effective dpi that the artists are rasterized at (by default it as at 72 dpi).