Why the initial figure dimensions are wrong in this matplotlib script?

Question

This Python 3.13 script with Matplotlib 3.10.3 run on 1200x1600 display:

import matplotlib.pyplot as plt, numpy as np

fig, ax = plt.subplots()
plt.tight_layout(pad=0.01)
fig.canvas.manager.full_screen_toggle()
img = np.random.randint(5, size=(4, 3))

while True:
  ax.imshow(img)
  print(fig.get_size_inches(), *ax.get_xlim())
  plt.waitforbuttonpress()

initially outputs wrong dimensions [6.4 4.8] 0.0 1.0, but after a click anywhere on the figure

it outputs physically correct values [12. 15.58] -0.5 2.5. What is a way to get correct dimensions before the first click, i.e. at the first call to print(fig.get_size_inches(), *ax.get_xlim())? My goal is to get the number of display pixels per unit square of img, so if there is an alternative to ax.get_position().width*fig.get_size_inches()[0]*fig.dpi/(xMax-xMin), which doesn't suffer from this problem, I would like to know.

An ugly solution is to insert these lines before the loop:

ax.imshow(img)
plt.pause(0.1)

Is there a better option?

Did you try fig.canvas.draw(), which is supposed to force layout computation without showing the figure? See also matplotlib's brief documentation. — JohanC
– JohanC, Commented Nov 18 at 9:03
@JohanC Yes, I tried with and without preceding ax.imshow(img) and it didn't work. — Paul Jurczak
– Paul Jurczak, Commented Nov 18 at 9:04
is that the most minimal example for which does the issue manifest? — cards
– cards, Commented Nov 18 at 15:33
@cards Depends on how you define minimal. You can reproduce it with a simple line plot, eliminating numpy dependence. — Paul Jurczak
– Paul Jurczak, Commented Nov 18 at 23:44

Chip01 · Accepted Answer · 2025-11-18 16:54:58Z

0

If I had to guess, I’d call plt.show(block=False) followed by plt.pause(0) before your first measurement.

This forces the GUI backend to finalize the window and apply fullscreen, so fig.get_size_inches() is correct immediately.

fig.canvas.manager.full_screen_toggle()

plt.show(block=False)

plt.pause(0)   # forces one GUI event cycle

The figure does not have its true on-screen size until the GUI event loop runs once. Fullscreen mode is only applied during that first event cycle. plt.pause(0) is the clean way to let the backend process pending events without adding delays, so the correct size is available at the first measurement.

(PS: All comments are appreciated, I’m new around here)

answered Nov 18 at 16:54

Chip01

1216 bronze badges

Sign up to request clarification or add additional context in comments.

4 Comments

Paul Jurczak Nov 19 at 7:07

After adding these lines, the script does not proceed to the while loop.

Paul Jurczak Nov 19 at 7:19

plt.pause(0.001) produces wrong dimensions, plt.pause(0.1) produces correct dimensions on my PC.

Chip01 Nov 19 at 13:51

Out of interest, what about plt.pause(0.01)?

Paul Jurczak 2 days ago

The same as plt.pause(0.001) - produces wrong dimensions.

Collectives™ on Stack Overflow

Why the initial figure dimensions are wrong in this matplotlib script?

1 Answer 1

4 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

4 Comments

Your Answer

Sign up or log in

Post as a guest

Related