Comments and Ratings

When I run mex DeleteGLTree.cpp on version R2009a with Mac OS 10.6 (Snow Leopard), I get the following message:

Undefined symbols:
  "_mexFunction", referenced from:
     -exported_symbols_list command line option
ld: symbol(s) not found
collect2: ld returned 1 exit status

    mex: link of ' "DeleteGLTree.mexmaci"' failed.

This turns out to be a problem with the declaration of mexFunction. For the solution, see
http://www.mathworks.co.kr/matlabcentral/newsreader/view_thread/173095

Send the data on my e-mail I need to see them.

Useful fantastic work.

Thanks Luigi, cool function - very useful for me!

You are right, the bug is generated only when the user choose to return distances. This is why I didn't realize before.

Thank you very much

Hi Luigi,

i really appreciate your tough work already for the 2d version.

However, there is some bug in your code that results in wrong results whenever the number of query points q does not match the number of data points p. If one adds a number of zeros to make the point numbers match, the code runs correct.

Are you able to fix this?

Best regards,
Johannes

I am sorry there is no paper.

Hello,
Thank you for the code.
The BuildGLTree.m file needs some slight corrections for this 3D version.
Best,
RK

just a niggle, it's profit, instead if profict.
I always "gambled" by trading stocks, but i think i will switch to poker :).
 

Uh,
I am sorry, those lines are for debug purpose, I forgot to delete them.
A "clean" version is on the way.

Outstanding! Very nice work.

cheese

Luigi provided a customized 3d one thats works great! thank you.

Given:
 
 N=reference +query points

Complexity of the algorithm is linear for random uniformly spaced points. Both for the serach and the tree construction.

 It becomes N^2 (Brute search) for very sparse dataset.

However this patological case can be detected after 5-10% running time. In this case is better to switch to kd-tree data structure. ( this detection is not present in this version)

                    Luigi

Please report questions on my e-mail, this space is dedicated to comments and ratings.

If you have a filled colud there are two main options, depending if your cloud comes from a tetraedral FEM mesh or not.

-If you have tetraedrons, than it is easy just get the boundary triangles.

-If you do not have them, it is more complicated, a ball pivoting or alpha shape tool is needed.

For more info, questions, suggestions contact me on my e-mail.

thanks a lot

Thanks a lot man...very good

It is very nice code for surface reconstruction.I find it very useful!!!Thx a lot.

Hello,

Thanks a lot.

Btw, do you have any ideas how to write codes for surface reconstruction using Fourier Transform Profilometry method (parallel fringe projection) I've taken images of reference fringe pattern and object on the pattern from digital camera.

Thanks in advance,

Regards,
Dave

I have not tested this for speed compared to the looped solution, but it is elegant. It is what I use in blktridiag.

%% Sparse block diagonal matrix, with no explicit loops
% Consider this matrix:
m = 4;
nblocks = 5;
A = rand(m,m*nblocks);

% We wish to generate a sparse block diagonal array,
% with five 4x4 blocks
[ind1,ind2,ind3] = ndgrid(0:m-1,0:m-1,0:nblocks-1);
rind = 1+ind1(:)+m*ind3(:);
cind = 1+ind2(:)+m*ind3(:);

% build the final array all in one call to sparse
Abd = sparse(rind,cind,A(:),nblocks*m,nblocks*m);

% Use spy to see the pattern is correct
spy(Abd)

%% Now do the same thing, but with 4x3 blocks
% Consider this matrix:
p = 4;
q = 3;
nblocks = 5;
A = rand(p,q,nblocks);

% use ndgrid
[ind1,ind2,ind3] = ndgrid(0:p-1,0:q-1,0:nblocks-1);
rind = 1+ind1(:)+p*ind3(:);
cind = 1+ind2(:)+q*ind3(:);

% build the final array all in one call to sparse
Abd = sparse(rind,cind,A(:),nblocks*p,nblocks*q);

% Use spy to see the pattern is correct
spy(Abd)

Pretty good. I like the help, example, etc.

For those interested, the sparse block diagonal matrix is built in a reasonable way. While it uses a loop the could be replaced with a call to ndgrid (see my own blktridiag for a mildly different way to build such a system) the loop here is a simple one that should not be costly.

I also like that the code is batched, so the matrices are limited in size, even for huge sets of problems to solve.

The only thing I'd have done differently would be to supply the matrix systems as a 3-d array, rather than concatenating them all horizontally. Make it a PxQxN matrix, where each plane of the array is a separate system to solve. This way if P>Q, then the linear solves will be least squares problems, all solved in a batch.