Yan Zhang

To Aureline,

Please type 'computer' in matlab and let me know what the result is. The precompiled version only works for 32bit windows system, yours may be 64bit.

To Yusuke,

The toolbox contains a AOS solver for the PDE in the form of u_t = div( g |\nabla u| ). This is a challenging problem because using explicit finite difference method (FDM) can cause unstability if the time step is too large.

For the PDE you mentioned, simple explicit FDM such as
u_(k+1) = u_k + delta_t * div(V)
should be working no matter how large delta_t is (of course the accuracy will be affected). The above scheme can be easily implemented in Matlab.

Hope this can help.

Hi Alex,

The problem has been fully discussed before, please see the previous comments for the details. In brief, go to the source file ac_distance_transform_3d.cpp, and get rid of the string in std::exception.

Hi Yan,

great code. I have one suggestion though: In your CV reinit process, you implemented a very efficient distance transform. However, for performance reasons, you don't compute the values for the edges of the matrix (always 0). Thus I would suggest, you pad the input matrix with 0s and then un-pad the output matrix of this step:

u0 = y_binary_boundary_detection(uint8(u>0));
u0 = padarray(u0, [1 1 1]); % padding
u0 = ac_distance_transform_3d(u0);
u0 = u0(2:end-1, 2:end-1, 2:end-1); % un-padding
u = u0.*sign(u);

Cheers, Christian