can i speed up this code and get rid of nested for loop??

 %coeffs1 is n*3 matrix with unknown n,i have no idea about it through 
 processing%%
 for k=1:length(coeffs1) 
   for i=1:length(unique(l))
      [r2, c2]= find(l == i & pd<inf);  % pd and l are matrices of 300*300 
      [r3, ~]=find(l==i & nc<inf); % nc is matrix of 300*300
          if ~isempty(r2)          
              v=zeros(length(r2),1);
              for n=1:length(r2)
                  v(n)=pd(r2(n),c2(n));
              end
              zbar=eval2dPoly(r2,c2,coeffs1(k,:)'); % Zbar is estimated new values using function eval2dPoly %%%%
              m=abs(zbar-v);
              r1=length(find(m==0));
              if r1/ length(r3)>=0
                  lamda=10;%10
                  dc=lamda*m*exp(-r1/length(r3));
                  Dc(k,i)=sum(dc);
              end
          end
    end
end

[Geoff Hayes]

Doaa - in your above code you have

 unique(l)

what is the variable l? I see that it is a 300x300 matrix, but is the data type integers or doubles or..? Also, please describe what you are attempting to do with this code; add comments where appropriate because it is unclear (to me) how this might be related to image processing or computer vision.

[Doaa Adel]

l: integer, labels of segmentation

it is an optimization problem over disparity map in segment domain, i have cross check map full of outliers, so i fit linear model over segments to get initial labeling for disparity map using these "linear planes".

then, i search for the best label for occluded portion, so i measure the cost between initial label and after having specific plane coefficients

   for k=1:length(coeffs1) 
   for i=1:length(unique(l))
      [r2, c2]= find(l == i & pd<inf);  % pd(cross check map)and l are matrices of 300*300 
      [r3, ~]=find(l==i & nc<inf); % nc (non occluded map)  matrix of 300*300
          if ~isempty(r2)          
              v=zeros(length(r2),1);
              for n=1:length(r2)
                  v(n)=pd(r2(n),c2(n));
              end
              zbar=eval2dPoly(r2,c2,coeffs1(k,:)'); % Zbar is estimated new values using function eval2dPoly which use the coeffs to find the estimated disparity values according to this coeff %%%%
              m=abs(zbar-v);
              r1=length(find(m==0));
              if r1/ length(r3)>=0
                  lamda=10;
                  dc=lamda*m*exp(-r1/length(r3));
                  Dc(k,i)=sum(dc);
              end
          end
    end
end

[Geoff Hayes]

How long does it take to execute your code given a typical input set? Seconds, minutes, hours?

Have you used the profiler to determine where the code is spending much of its time? Try using this to see where you might be able to improve upon your code.

Also, consider attaching mat files of your data so that others can run through the code and get an idea of what is happening. This mat file should include (at least) the coeffs1, l, nc, and pd variables.

[per isakson]

... and preallocate Dc

[Doaa Adel]

thank you Geoff Hayes, i use profiler always, this part of code takes about 1600 seconds and that is so so long.

thank you per isakson, of course i did that but it is just not mentioned here

[per isakson]

It's difficult to help without running the code, which is difficult without input data.

What's make you believe that it is the for-loops, which cause the problem?