CGAL users discussion list

[Bryn Lloyd <>]

Hi Sebastien

Thanks for your reply.

Sharing my code is a bit difficult, since it getting quite large (custom 
label mesh domain derived from Labeled_mesh_domain_3, custom facet 
sizing field image computed at same resolution as segmentation, etc.). 
To clarify: I mean the smoothness of the domain interface surfaces.

What I observed so far:
- It seems to make the domain interfaces smoother if I switch on ODT and 
Lloyd optimization. However, I wonder why the cropped segmentation is so 
much smoother, even without ODT/Lloyd optimization.

- Making the facet distance smaller does not seem to make domain 
interfaces smoother.

- The noisy domain interfaces occur where the mesh is finest, where I am 
trying to resolve thin skin-like tissue regions.

Questions:
- Could it be related perturbation/exudation (e.g. default time limits)?

- What other parameters should I try?

- Is my strategy reasonable to define a non-constant facet sizing field, 
and a constant cell sizing and constant facet distance? Do you have any 
recommendations?

- I pass a Bbox to the mesh domain? For the full body segmentation this 
box is long in z-direction, for the cropped region the box is nearly a 
cube. Could this have an influence? How are the points sampled during 
mesh generation?

Thanks for any suggestions!
Bryn




In general,

On 22.05.2015 11:04, Sebastien Loriot (GeometryFactory) wrote:
> Could you share the code used?
> It might be that you have a parameter that is depending on the bbox of 
> the model (which would explain the smoothness when the model's bbox is
> smaller).
>
> Sebastien.
>
> On 05/21/2015 03:00 PM, Bryn Lloyd wrote:
> > Hi
> >
> > I am using CGAL::make_mesh_3 with CGAL::Labeled_image_mesh_domain_3 and
> > a sizing field for the facets.
> >
> > When I generate a mesh of a full anatomical body segmentation it is very
> > spiky/noisy.
> > When I generate a mesh using the same parameters, but a subset of the
> > segmented image (1/4 in longest axis), the mesh is much smoother.
> > Why does the mesh change for exactly the same model and parameters? Has
> > anybody had any similar problems? What could help to improve the mesh of
> > the full body?
> >
> > Note:
> > - My image has voxels with spacing 0.5, my sizing field is in the range
> > [0.5, 6].
> > - Other parameters include: Mesh_criteria criteria(facet_angle=25,
> > facet_size=facet_sizing_field, facet_distance=1.0,
> > cell_radius_edge_ratio=2.0, cell_size=Constant_sizing(6.0) );
> > - I run the make_mesh_3 command with default args (without ODT and Lloyd
> > smoothing, but with perturb/exude ON).
> >
> > Any suggestions/help would be greatly appreciated.
> >
> > How are the mesh points sampled or domain intersections found? Does the
> > bounding box (shape) of the domain/oracle have any influence?
> >
> > Thanks, Bryn