Overall, this is a very useful package, which an exceptionally good documentation (See PDF file!) Its main advantage are generic, well accessible interface and the modular, integrated workflow (see the Fig.2 in the manual).
For me (MATLAB 2012b, ubuntu64bit), it works out of the box, and automatically compiled the missing mex files for your platform.
However, I encountered a few Out-Of-Memory issues with the "smoothing" option. If you set the default value,
p.smoothing.smooth = false; (ExportVoxelData.m, line 137) then it works. The "geometric" option took very long, I suppose it got stuck in CONVERT_meshformat.
I think, if you have 3D-data e.g. from X-Ray-Tomography or confocal microscopy, and are looking for nice visualization -even with stereoscopic images- of segmented objects in Povray or with your favourite STL program, this might be the right tool.
Thanks for providing this useful piece of numerical optimization code. I implemented it for fitting tomographic data to a model. I found that the step sizes for computing the gradient (DiffMinChange) seemed quite crucial for the convergence. Since my set of variables for optimization has different domains (e.g. [0...0.5] or [0....2*PI] ...), I had to modify the code so to account for different DiffMin/MaxChanges.
BTW, is there a fast MEX-version of the (L)BFGS optimizer?
This method works quite well for binarizing my samples!
Note: I found that, depending on the grayvalue variations, a lot of background pixels are wrongly segmented as foreground. However, masking with another binary image, binarized e.g. with Otsu's threshold (function graythresh), remedied this.