Showing np.uint16 images of the form (h,w,3) is broken

[luispedro]

Here is a simple test case:

import matplotlib.pyplot as plt
import numpy as np
A = np.array([np.arange(256,dtype=np.uint16) for _ in xrange(256)])
B =  (A<<8)+(255*np.random.random(A.shape)).astype(np.uint16)
B = np.dstack([B,B,B])
plt.imshow(B)
plt.show()

This results in random bits.

The problem is in file lib/matplotlib.cm.py: the conversion to np.uint8 assumes that the input is in 0..1 format.

[tacaswell]

To be fair, the docs (http://matplotlib.org/api/axes_api.html#matplotlib.axes.Axes.imshow) only claim to support float or uint8 so I would not call this broken.

[luispedro]

Then, maybe the real bug is that it doesn't raise an exception :)

It's certainly surprising behaviour.

[tacaswell]

@luispedro Can you make a PR with that fix?