CGAL users discussion list

[Stephane Tayeb <>]

Hi,

Dominik Szczerba wrote:
> Hi Stephane,
>
> > > 1) on the provided mesh_3D_image example of the liver: changing size 
> > > and approximation criteria has no effect on the output mesh
> > Could you give more details please ? For example can you see any 
> > difference between
> > Facet_criteria(30,6,1) and Facet_criteria(30,6,4) ?
>
> I see almost not difference between e.g. Facet_criteria(30,6,4) and 
> Facet_criteria(30,6,10). But ineed, if I experiment more also with the 
> other parameters I seem to be getting the rules.
Sometimes some parameters are stronger than others. It may be the case 
here. Perhaps size criterion (which is set to 6) is more restrictive 
than approximation criterion for values > 4.

> > > 2) in the liver example, the boundaries are very noisy; the 
> > > difference to just interpolating the labels onto a non-conformally 
> > > generated mesh is not striking.
> > The edges are noisy (they are not handled by Mesh_3, for now). 
> > Surfaces should be ok, if you don't try to over approximate them. The 
> > image grid will be visible if the mesh resolution is too high.
>
> True, with smaller resolution the surface is smoother, thanks for a 
> hint. However, I start losing features then. A curvature/proximity 
> refinement rules would be good to have here. But I guess this is just 
> a feature of this method for now.
I'm not sure to understand. The approximation error is a curvature 
refinement rule. It is a bound on the distance between the center of the 
surface facet (i.e. center of the circumcenter of the triangle) and 
intersection point between the voronoi edge of this facet and the surface.
> > > 3) the labels in the output seem shuffled: they do not seem to 
> > > correspond properly to the image label information. It means I can 
> > > not identify my domains from the original label tags.
> > This is right. Subdomain (int) and surface indices 
> > (std::pair<int,int>) are mapped to int when writing the output. For 
> > example, if subdomains are {1,7,12} and surfaces are {<1,7>, <7,12>, 
> > <0,12>} (0 is the outside), medit writer may remap as follows :
> > 1 -> 1
> > 12 -> 2
> > 7 -> 3
> > <1,7> -> 4
> > <7,12> -> 5
> > <0,12> -> 6
> >
> > There is a flag (CGAL_MESH_3_IO_VERBOSE) which makes medit writer 
> > verbose. This may help you.
>
> Thanks, but I would actually like to 1) preserve my original IDs (I 
> maintain a global map integer<->string) and 2) bypass the stony medit 
> format and save my mesh directly into hdf5 or at least compressed XML 
> VTK (I have detailed whole body models here and can't afford any 
> ASCII). Could you please point me to an example how I can retrieve the 
> mesh AND the ORIGINAL labels? As I can see from the mailing list 
> archives I do not seem to be the only person puzzled by this seemingly 
> simple task... :)
c3t3 object describes the mesh and allows to get the mesh and the 
original labels. Actually, original labels are stored into c3t3 as 
subdomain indices of cells (note that they may be cast from unsigned 
char to int). And surface indices of facets are a pair of two original 
labels (e.g. if facet is between subdomain 1 and 2, it stores <1,2> as 
surface index).

You can find in MeshComplex_3InTriangulation3 documentation page how to 
access these indices. Or you can take a look in CGAL/IO/File_medit.h to 
see how it is done to write in medit format.

Regards,

--
Stephane Tayeb
Software engineer - INRIA Sophia Antipolis
Geometrica Project-Team