Python - More efficient way to loop through matrices

Question

I have

an (height x width x 3) RGB image img
a (256 x 256 x 256) matrix color_mapping mapping colors
a (height x width) matrix pixel_result which I want to fill with the elements mapped with color_mapping for each pixel of the image

How can I fill that last matrix efficiently?

So far, I simply run through all the pixels in for loops, but I'm wondering if there is a more efficient way to do it

for i in range(width):
    for j in range(height):
        pixel_result[j,i] = color_mapping[img[j,i,0],img[j,i,1],img[j,i,2]]

willeM_ Van Onsem · Accepted Answer · 2020-02-29 10:32:41Z

1

You can implement this by subscripting the color_mapping:

pixel_result = color_mapping[img[:,:,0], img[:,:,1], img[:,:,2]]

For example if test the performance with:

>>> def f():
...   for i in range(width):
...     for j in range(height):
...         pixel_result[j,i] = color_mapping[img[j,i,0],img[j,i,1],img[j,i,2]]
... 
>>> def g():
...     pixel_result = color_mapping[img[:,:,0], img[:,:,1], img[:,:,2]]
>>> color_mapping = np.random.randn(256,256,256)

Then for a small image: img = np.random.randint(0, 256, (12, 7, 3)), we get:

>>> timeit(f, number=10000)
0.5247259399620816
>>> timeit(g, number=10000)
0.032307324931025505

For larger images, like img = np.random.randint(0, 256, (1920, 1080, 3)), we get:

>>> timeit(f, number=10)
18.934733690926805
>>> timeit(g, number=10)
0.5807857210747898

or for a 72×128 image:

>>> timeit(f, number=100)
0.6014469779329374
>>> timeit(g, number=100)
0.011570235947147012

edited Feb 29, 2020 at 10:32

answered Feb 29, 2020 at 10:07

willeM_ Van Onsem

482k33 gold badges483 silver badges624 bronze badges

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

6 Comments

user3548298 Over a year ago

It's actually 15 times slower

willeM_ Van Onsem Over a year ago

@user3548298: the assignment is of course done once, so not in a for loop. I've tested this for both small and large images, and got a speedup of approximately 32 for large images, and 16 for small images.

user3548298 Over a year ago

Oh ... right I'm an idiot... I forgot to take it out of the loop

willeM_ Van Onsem Over a year ago

@user3548298: you noticed that there are two loops that you need to remove? :) I got a speedup of 52x for that.

willeM_ Van Onsem Over a year ago

@user3548298: not with numpy. Here I used floating point values as well (notice that np.random.randn(..)). You can try to use a tool like cudamap to process this on a GPU.

|

Collectives™ on Stack Overflow

Python - More efficient way to loop through matrices

1 Answer 1

6 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

6 Comments

Your Answer

Sign up or log in

Post as a guest

Related