Bug report

Bug summary

When using multiple calls to plt.imshow() with transparancy/alpha on the top image, the .pdf and .png versions are inconsistent

Code for reproduction

Code for reproduction below, also available as a binder at https://mybinder.org/v2/gist/TAdeJong/8cb84dc74bc6d2c2a1744ef465bf85d8/HEAD .
Effect can be even more noticable when overlaying actual images, but for simplicity this example keeps it a single colors:

import matplotlib.pyplot as plt
import matplotlib
import numpy as np

print(matplotlib.__version__, matplotlib.get_backend())

fig,ax = plt.subplots(ncols=11, figsize=[12,3])
for i in range(11):
    ax[i].imshow(np.full((100,100), 0.1), cmap='gray', vmin=0, vmax=1)
    ax[i].imshow(np.full((100,100), 0.9), cmap='inferno', vmin=0, vmax=1, alpha=0.1*i)
    ax[i].set_title(f"$\\alpha=${0.1*i:.1f}")
plt.tight_layout()
plt.savefig('test.png')
plt.savefig('test.pdf')

Actual outcome

Output files:

Colors in output files seem consistent between different OS and viewers.

# Mybinder version info:
3.3.2 module://ipykernel.pylab.backend_inline
# Local version info:
3.3.2 TkAgg

Expected outcome

Produced colors being consistent between different output formats, i.e. a equal color for each of the respective squares between the PNG and PDF output.
OR a clear warning in the documentation (I am not aware of such a warning existing, but I might have missed it).

Matplotlib version

• Operating system: Ubuntu/mybinder
• Matplotlib version: 3.3.2, at least down to 3.2.2
• Matplotlib backend (print(matplotlib.get_backend())): module://ipykernel.pylab.backend_inline / TkAgg
• Python version: 3.7.7, 3.7.3, 3.7.8
• Jupyter version (if applicable):
• Other libraries:

#Edit: fix mybinder link.