Are Image Operations just Slow?

Technical Support Python Support
1

I’m trying to implement image convolution in a blender script (for say, edge detection).

Firstly I’m trying to separate a single colour channel of a loaded image into an array 1/4 of the size to use in the convolution, however it seems painfully slow to do this:

import bpy
import math

D = bpy.data
image = D.images.load('C:/Blender Files/textures/peppers.jpg', check_existing=True)

def channel(img, c):
    buff = img.pixels
    chan = []
    chan = [0 for i in range(0, int(len(img.pixels)/4))]
    
    for i in range(0, int(len(img.pixels)/4)):
        if (i%1000 == 0):
            print(i)
        chan[i] = buff[4*i+c]
        
    print(chan)
    return chan

red = channel(image, 0)

It prints out 1000 every couple of seconds meaning it takes approximately 2ms to copy over a pixel value, is this just due to the nature of python scripts in blender or is there a faster way to do this?

Secondly, if it is this slow to just extract the red channel of the image array, is there any hope in writing a image convolution function that iterates over the image with much more calculation at each pixel?

2

Update:

So I read that accessing individual pixels of a loaded image can be slow, hence why I’d tried copying the array into a new buffer, buff. However buff = img.pixels just sets the buff to point to the original image object, buff = img.pixels[:] makes a duplicate.

This solved the problem quick time.

I guess my new question is now aimed at the devs;
if it takes 2ms to change or read a specific pixel object but almost instant to copy the whole array into a new buffer, why is there not an intermediate buffer used in the implementation of how image objects are accessed? and why is the nature of them so slow?

3

You can get the pixels of the image even faster by using foreach_get and foreach_set. Have a look here
https://developer.blender.org/rB9075ec8269e7cb029f4fab6c1289eb2f1ae2858a

4

Interesting, thank you!

5

related: Fill large image via Python