Help me to optimise or vectorize the code

I need to optimize this snippet of code where siz is close to 300. tried vectorization with NDGRID but it takes too much memory and matlab throws an error

finalMat = double(zeros(siz,siz,siz));
  for i = 1:size(point3d,1)
    for j = 1:size(point3d,2)
        finalMat(point3d(i,j,1),point3d(i,j,2),point3d(i,j,3))  = max(value(i,j),finalMat(point3d(i,j,1),point3d(i,j,2),point3d(i,j,3)));
    end
  end

[Adam]

Isn't that code just storing value(i,j) if it is positive and 0 otherwise or am I missing something?

[Ank]

No basically I am converting the 3d coordinate stored in point3d into to actual 3d array (finalMat). the index might repeat so taking the maximum value of all such occurance.

[Joseph Cheng]

I'm curious that it takes too much memory (edit: just re-read and realized siz is not apply to size of point3d. What is the size of point3d?) and what the error is. What is the error it gives? is point3d contain non-integer numbers?

point3d = imread('pears.png');
value = rgb2gray(point3d);
siz = 300;
finalMat = double(zeros(siz,siz,siz));
tic
  for i = 1:size(point3d,1)
    for j = 1:size(point3d,2)
        finalMat(point3d(i,j,1),point3d(i,j,2),point3d(i,j,3))  = max(value(i,j),finalMat(point3d(i,j,1),point3d(i,j,2),point3d(i,j,3)));
    end
  end
  toc

I tried the above code as a vague try to experiment with it and it goes quickly and doesn't throw an error.

[Ank]

Currently it gets done in 100 milliseconds but i was looking for it to be faster. Size of point3d is is like 2000x2000.

Answer 1

[Guillaume]

You could try using accumarray, but I'm not sure it'd be any faster:

rowsubs = point3d(:, :, 1);
colsubs = point3d(:, :, 2);
pagsubs = point3d(:, :, 3);
finalMat = accumarray({rowsubs(:) colsubs(:) pagsubs(:)}, value(:), [siz siz siz], @max);

If you can modify your algorithm to store your point3d coordinates and your values as vectors, you could dispense with all the (:) in accumarray.