Avoid using repmat in CPD registration algorithm

Hi all,

I'm running a coherent point drift (CPD) registration algorithm and I'm having trouble running it on large point clouds because I keep exceeding the maximum variable size. The line of code that is giving me trouble is:

G=repmat(x,[1 1 m])-permute(repmat(y,[1 1 n]),[3 2 1]);

The error message is:

??? Maximum variable size allowed by the program is exceeded.

Error in ==> repmat at 104
    B = A(subs{:});

I'm wondering if anyone has any suggestions as how to avoid using repmat so I can run the code on large data sets.

Thanks!

[Matt J]

Perhaps. What are the dimensions of x and y?

[Mark]

I'm not exactly sure as there can be significant deviation in the point clouds. Let's say the largest X will be is 70000 x 3 and the largest Y will be is 4000 x 3.

Answer 1

[Sean de Wolski]

bsxfun should be able to handle that.

Also, rather than using permute use reshape, permute is slow for big things.

G = bsxfun(@minus,x,reshape(y',1,3,[]))

[Mark]

Thanks for the suggestion Sean. I'll try it out and let you know if it works!