I'm witnessing a behavior that seems like a bug when using imshow() with the interpolation flag. Consider this example:

from matplotlib import pyplot as plt
W=np.random.randn(200,200)
plt.imshow(W, interpolation='bicubic', cmap='seismic', vmin=-3, vmax=3)

This results in the following image:

Obviously I did not expect the squares to be there. This issue also happens with the option bilinear, gaussian, bessel and pretty much every other interpolation that I tested that isn't nearest.

Here's the output of the nearest option, just for comparison:

So this is weird in that the squares are larger than a "data pixel". I don't know what's causing this. The squares apparently pop where the value is larger/smaller than vmax/vmin, but that's about all I know.

Some notes:

• If I don't set vmin and vmax this effect doesn't happen.
• I'm using version 2.0.2 of matplotlib
• This issue happens in Python 3.4 and 3.5 and Python 2.7 with said version of matplotlib
• Doesn't appear to affect Python 2.7 and matplotlib 2.0.0