0

I have the following nested loop which goes over a big image (7020x5100) and sets pixel values of another image with the same dimension depending of the value of i. The problem that this is very very slow...could you please give a hint, how to speed up this algorithm? 

A=[]

for x in xrange(x_img):
    for y in xrange(y_img):
        A.append(probImage1[x][y])
        A.append(probImage2[x][y])
        A.append(probImage3[x][y])
        i = np.argmax(A)
        if i == 0:
            processedImage[x,y] = [1, 89, 255]
        if i == 1:
            processedImage[x,y] = [241, 28, 3]
        if i == 2:
            processedImage[x,y] = [137, 254, 255]
        A = []
Kristan
  • 387
  • 6
  • 19
  • Try to use inbuilt numpy methods instead of your own loop – ZdaR Jan 19 '17 at 09:46
  • Yes, but it is quite "complex" since the value i has to determined in each step, and I do not know how to do it another way... – Kristan Jan 19 '17 at 10:10
  • That code looks incomplete. Hard to deduce if you can calculate the A.append stuff just outside a-priori. This beeing done it's simple to write the values using numpy's vectorized formulations. – sascha Jan 19 '17 at 11:53

1 Answers1

1

On my machine your method takes 6 mins 55 seconds!

Try to avoid for loops iterating over an image, especially with all memory allocation going on. This method takes 2.14 seconds for a 7020 x 5100 image

newProb=np.dstack((probImage1,probImage2,probImage3))
A=np.argmax(newProb,axis=2)
processedImage[A==0]=[1,89,255]
processedImage[A==1]=[241,28,3]
processedImage[A==2]=[137,254,255]
docPhil
  • 190
  • 8