CGAL users discussion list

[Sebastian Bachmann <>]

On 09.05.2022 16:13, Sebastien Loriot ( via 
cgal-discuss Mailing List) wrote:
> On the attached screenshot, you'll see in read the intersection of
> the output mesh with the cube (0 0 0 -> 100 100 100).
>
> So clearly, the mesh is not exactly on the cube.
> The initialization (intersection of the image with the cube) is
> done with points exactly on the cube. But then, intersection points
> are computed by bisection (it is handled as an image).

okay I see. but in that case I should see some points directly on the 
image boundary? Or does this not apply to the mesh vertices itself?

> Snapping could be done by using a dedicated oracle using a different
> intersection computation function when the segment intersect the
> boundary cube.

Do I understand it correctly, that the bisection is handled here: 
https://github.com/CGAL/cgal/blob/acbd1ee182211c6309f355925927802f036fbe13/Mesh_3/include/CGAL/Labeled_mesh_domain_3.h#L713
?

I.e. I would have to change that, such that in the case of the image 
boundary, not `mid` is returned but a point exactly on that surface?

Best,
Sebastian



> Best,
>
> Sebastien.
>
>
> On 5/9/22 10:43, hello ( via cgal-discuss Mailing List) wrote:
> > Hello,
> >
> > sure! Attached is an example source code and the resulting file for 
> > the 420 example.
> >
> > I would have expected, that all nodes on the bounding surfaces are 
> > either at x/y/z 0 or 100 (as it is for the edges) but instead they 
> > appear roughly at 0 or 100 +-0.05.
> > It seems to correlate with the relative_error_bound, however I can 
> > only make the error smaller but not go away completely. (The attached 
> > file is generated with 1e-3)
> >
> > Best,
> > Sebastian
> >
> >
> > On 09.05.22 09:23, Sebastien Loriot ( via 
> > cgal-discuss Mailing List) wrote:
> > > Hello,
> > >
> > > could you share a minimal example showing the problem?
> > >
> > > Thanks,
> > >
> > > Sebastien.
> > >
> > > On 5/7/22 13:37, Sebastian Bachmann ( via cgal-discuss 
> > > Mailing List) wrote:
> > > > Hello,
> > > > No one has an idea? Is it not somehow possible to restrict nodes 
> > > > onto a surface? I.e. if I have the polyline of the surface's 
> > > > boundary, can I specify that all nodes must be on that surface?
> > > >
> > > > For now I have written a function to go over all nodes and check if 
> > > > they are near such a boundary face and then "pull" the nodes onto 
> > > > that surface. However, there are certain edge cases were this then 
> > > > yields collapsed triangles etc. Thus, solving this during meshing 
> > > > would be a great enhancement.
> > > >
> > > > Best,
> > > > Sebastian
> > > >
> > > > On 25.04.2022 07:16, Sebastian Bachmann ( via 
> > > > cgal-discuss Mailing List) wrote:
> > > > > Hello,
> > > > >
> > > > > I created an image meshing tool, similar to the one given here in 
> > > > > this example code: 
> > > > > https://doc.cgal.org/latest/Mesh_3/Mesh_3_2mesh_3D_image_with_features_8cpp-example.html 
> > > > > with the only difference, that I extract the surface elements and 
> > > > > not write the full 3D mesh.
> > > > > I noticed, that on the image boundary, the vertices would be not 
> > > > > exactly on the face but roughly +- the relative error bound given 
> > > > > in create_labeled_image_mesh_domain. For examples, nodes on the xy 
> > > > > plane at z=0 are not on z=0 but for example -0.01 to 0.01. While 
> > > > > the protected polylines are exact and have always z=0.
> > > > > I created for example a mesh from the 420.inr example with the same 
> > > > > meshing criteria as given in the example and can see that behaviour 
> > > > > there.
> > > > >
> > > > > While you can control the maximal deviation using the relative 
> > > > > error bound, there will always be some deviation. And if I set the 
> > > > > relative error bound too low, it seems the meshing never finishes.
> > > > > I wonder now, if it is possible to constraint the mesh (similar to 
> > > > > the 1d polylines), such that the vertices for facets on the image 
> > > > > boundary, are always exactly on the image boundary?
> > > > >
> > > > > Best,
> > > > > Sebastian