Minimal code for rendering a numpy array as an image in a Jupyter notebook in memory. Borrowed from the Deep Dream notebook.

showarray.py

import PIL.Image
from cStringIO import StringIO
import IPython.display
import numpy as np
def showarray(a, fmt='png'):
    a = np.uint8(a)
    f = StringIO()
    PIL.Image.fromarray(a).save(f, fmt)
    IPython.display.display(IPython.display.Image(data=f.getvalue()))

IPython can display PIL.Image directly! So you can skip the encode-to-png and decode-from-png:

IPython.display.display(PIL.Image.fromarray(a))

Actually,

PIL.Image.fromarray(a)

in jupyter notebook, is enough.

Thanks

things can be made easier if you have cv2: https://gist.github.com/ctmakro/3ae3cd9538390b706820cd01dac6861f

@ctmakro - thank you for the elegant solution -- image shows directly in Jupyter Lab 👍

This works for Google colab (mostly for my own reference)

import IPython.display
import PIL.Image
IPython.display.display(PIL.Image.fromarray(a))

It's a lot simpler with matplotlib:

from matplotlib.pyplot import imshow
%matplotlib inline

#image is a numpy array
imshow(image)

https://gist.github.com/prampey/dc0e3326131baaf35c380e7ddbf5119a

imshow does not faithfully render an array as an image and is generally annoying. There is good reason to use PIL for this

That’s strange, Imshow Has worked for me with any 3D NumPy array. Do you have an example? @rueberger
In fact, it also works with PIL images out-of-the-box.

import PIL.Image
imshow(PIL.Image.open(img_path))

It interpolates by default. From the docs "The number of pixels used to render an image is set by the Axes size and the dpi of the figure. This can lead to aliasing artifacts when the image is resampled because the displayed image size will usually not match the size of X".

Just a lot of nonsense to wade through when you just want to see a faithful representation of an image array