Transforming python for-for loop for accessing pixels into multithreaded python code or GPU-threaded python

Question

Supposing I have to run this code for 500.000.000 pixels.

import cv2

path_of_the_picture = "photo.jpg"

picture = cv2.imread(path_of_the_picture, cv2.IMREAD_UNCHANGED) 

number_of_rows,number_of_cols = picture.shape

for x in range(number_of_rows):
    for y in range(number_of_cols):
        picture[x,y] = picture[x,y] + 10

It takes around 30min to complete! Ok, maybe it needs some editing to run, but you get the idea. How can I accelerate the above code with multithreaded code in python? Moreover, can I use some GPU-CUDA power? After searching I found this suggestion: How to Multi-thread an Operation Within a Loop in Python but it does not work for my situation...

what is the actual opperation ? it's hard to know which solution adapts better to your problem without knowing the actual problem... here, you could convert image to numpy array and then add 10: picture = np.array(picture) + 10 and i'm sure this is fast enough — Ulises Bussi
– Ulises Bussi, Commented Jun 27, 2022 at 15:16

Ulises Bussi · Accepted Answer · 2022-06-27 15:45:34Z

1

Here you could use numba to speed up the solution. But I think that not knowing you exact problem is hard to estimate best solution:


import numba as nb
import numpy as np
from time import time

n=22400 #aproximately 5e8 pixels
a =np.random.randint(0,255,(n,n),dtype=np.uint8)

def func1(a):
    n=len(a)
    for x in range(n):
        for y in range(n):
            a[x,y] = a[x,y]+10
    return a

@nb.jit
def func2(a):
    n=len(a)
    for x in range(n):
        for y in range(n):
            a[x,y] = a[x,y]+10
    return a


t0 = time()
q1 = func1(a)
t1 = time()
q2 = func2(a)
t2 = time()
q3 = a+10
t3 = time()
print("func1 takes: ", t1-t0)
print("func2 takes: ", t2-t1)
print("numpy takes: ", t3-t2)

Results:

#wich outputs in seconds:
#func1 takes:  1008.4661099910736
#func2 takes:  0.15050935745239258
#numpy takes:  0.1150052547454834

answered Jun 27, 2022 at 15:45

Ulises Bussi

1,7351 gold badge5 silver badges18 bronze badges

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

4 Comments

nobody Over a year ago

Here: a =np.random.randint(0,255,(n,n),dtype=np.uint8) how do I write it, since I know rows and columns so that I do not need random?

Ulises Bussi Over a year ago

here a is an example matrix just to run the code... because I DON'T HAVE THE ORIGINAL FILE TO WORK, it was not proportioned in the post. so I invented one "image". just add the @nb.jit to your original code...

nobody Over a year ago

So, there is no need to use numpy ?

Ulises Bussi Over a year ago

no, just numba, or you can use numpy to solution 3 (transform to array and add 10) there use picture = np.array(picture )+10

Collectives™ on Stack Overflow

Transforming python for-for loop for accessing pixels into multithreaded python code or GPU-threaded python

1 Answer 1

4 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

4 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related