Why can't I assign uint16 array to array vector in Python?

Question

I run the following code:

    colored = np.empty((299, 299, 3), dtype=float) # dtype=`uint16` doesn't work too!
    colored[:, :, 0] = image
    colored[:, :, 1] = image
    colored[:, :, 2] = image

If I plot image (2 dimensional array of uint16), I can see the correct picture. But if I plot colored array then I get the following trash image:

I plot with the following code:

    import matplotlib.pyplot as plt
    plt.imshow(colored, cmap='gray')
    plt.show()

When I plot image it works. When I plot colored it doesn't. What's going on here and how can I fix that?

What plot command are you using? Do the numbers look right ? — hpaulj
– hpaulj, Commented Oct 7, 2017 at 21:04
I use plt.imshow(colored, cmap='gray'). Numbers look right! So, something's wrong with plot function... — Dmitry
– Dmitry, Commented Oct 7, 2017 at 21:08
But I'm trying to plot colored and it should be array of float. When I'm plotting image (array of uint16) by the same way, it works! — Dmitry
– Dmitry, Commented Oct 7, 2017 at 21:11
How about imshow(colored[:,:,0])? When creating the 3d array, are you normalizing the values. imshow says the values should be in the 0.0 to 1.0 range. — hpaulj
– hpaulj, Commented Oct 7, 2017 at 22:23
Both 2D array and 3D array of 0.0-1.0 values work fine. But I thought that imshow can be used with 3-color images with values 0..255... — Dmitry
– Dmitry, Commented Oct 7, 2017 at 23:22

unutbu · Accepted Answer · 2018-05-14 11:40:18Z

2

When the array passed to imshow is 3-dimensional and has floating-point dtype, plt.show() calls this piece of code:

if x.ndim == 3:
    ...    
    if xx.dtype.kind == 'f':
        if bytes:
            xx = (xx * 255).astype(np.uint8)
    return xx

So the values in colored are being multiplied by 255 and then astyped to np.uint8. This happens before the values are normalized to the interval [0, 1]. You can verify this by putting a print statement in your matplotlib/cm.py file:

if x.ndim == 3:
    ...    
    if xx.dtype.kind == 'f':
        if bytes:
            print(xx.max())
            xx = (xx * 255).astype(np.uint8)
    return xx

and running this script:

import numpy as np
import matplotlib.pyplot as plt

maxval = np.iinfo('uint16').max
uint16_image = np.linspace(0, maxval, (16**2)).astype('uint16')
uint16_image = uint16_image.reshape(16, 16)

colored = np.empty(uint16_image.shape + (3,), dtype=float)
colored[:, :, 0] = uint16_image
colored[:, :, 1] = uint16_image
colored[:, :, 2] = uint16_image

fig, ax = plt.subplots(nrows=2)
ax[0].imshow(uint16_image)
ax[1].imshow(colored)
plt.show()

which prints

65535.0
65535.0

and displays

The moral of the story is that if you pass a 3D float array to plt.imshow, the values must be between 0 and 1 (or NaN) or else you will likely get (graphical) mojibake.

edited May 14, 2018 at 11:40

answered Oct 8, 2017 at 3:26

unutbu

886k197 gold badges1.9k silver badges1.7k bronze badges

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

2 Comments

l'L'l Over a year ago

Using your example above, what would cause the second plot gradient to be blank aside from the black square in the upper left?

unutbu Over a year ago

The difference in behavior is caused by this change to the matplotlib source code. (See pull 10220). At this point in the code the dtype of self._A is float64 so the high is set to 1. self._A = np.clip(self._A, 0, high) now clips everything to 0 or 1.

Collectives™ on Stack Overflow

Why can't I assign uint16 array to array vector in Python?

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related