CGAL users discussion list

[Sebastian Bachmann <>]

Hello,

Yes, I was using exactly that code, but it is depending on the random 
number generator. I had to run it several times until such a hole would 
appear. I tried again now, and it took me 6 attempts.

I have uploaded a new version where I patched 
initialize_triangulation_from_labeled_image to use a specific random 
seed. In my case, 1234 triggered the problem.
I'm using Debian Bookworm.

Best,
Sebastian


Am 19.05.2025 um 09:00 schrieb Sebastien Loriot ( 
via cgal-discuss Mailing List):
> Hello,
>
> when running you minimal example I don't have the "holes" you are
> having. I'm also using CGAL 6.0.1. What platform are you on?
> Are you really using that example or a modified version?
>
> Best,
>
> Sebastien.
>
> On 4/28/25 10:54, Sebastian Bachmann ( via cgal-discuss 
> Mailing List) wrote:
> > Hello!
> >
> > Anyone has an idea what could cause that?
> > In like 95% of meshes I create I do not have that problem, but once in 
> > a while this turns up and is annoying...
> > It seems to be more often the case, when more than two labels are used 
> > (i.e, 0, 1 and 2 and not just 0 and 1).
> >
> > Anything that I could test will definitely help me :)
> >
> > Best,
> > Sebastian
> >
> > Am 18.02.2025 um 09:06 schrieb Sebastian Bachmann ( via 
> > cgal-discuss Mailing List):
> > > Hi!
> > >
> > > Sure, here is an example: https://files.reox.at/.mesh/
> > >
> > > You may have to run it several times until such a hole appears, as no 
> > > optimization is present. However, you can make them appear much 
> > > easier by changing the line with refine_mesh_3 to:
> > >    CGAL::refine_mesh_3(c3t3, domain, criteria, CGAL::parameters::odt());
> > >
> > > ODT seems to be more prone than Lloyd.
> > > I uploaded two surface.off files as well, where this can be seen 
> > > (surface.off is without odt/lloyd and the second run, surface_odt.off 
> > > is with ODT active).
> > >
> > > Best,
> > > Sebastian
> > >
> > > Am 17.02.2025 um 11:34 schrieb Jane Tournois:
> > > > Hello,
> > > > can you please post some code and data so that we can try to 
> > > > reproduce your issue?
> > > > Best,
> > > > Jane
> > > >
> > > > Le 07/02/2025 à 16:30, Sebastian Bachmann ( via cgal- 
> > > > discuss Mailing List) a écrit :
> > > > > Hi!
> > > > >
> > > > > I send this mail to the list yesterday, but apparently it was not 
> > > > > posted - but I also did not get a rejection mail? I suspect that 
> > > > > attachments are not allowed and get silently dropped?
> > > > > So here it is again with a link instead, sorry if it is now posted 
> > > > > twice!
> > > > >
> > > > >
> > > > > I extended the example code from 
> > > > > mesh_3D_image_with_custom_initialization.cpp to mesh a multi-label 
> > > > > 3D image, i.e., I have for example 3 voxel values: 0 (background), 
> > > > > 1 (some part), and 2 (some other part).
> > > > > This works reasonable well, but if I go down with the cell size, I 
> > > > > get weird artifacts in the surface of the mesh: at some random 
> > > > > places, "holes" appear (see screenshot here: https:// 
> > > > > files.reox.at/.mesh/ screenshot.png).
> > > > >
> > > > > I tried to track down the problem and it can be triggered by using 
> > > > > lloyd or odt optimization. When either of these are used, such 
> > > > > holes appear in my mesh (much much more in odt) - if I switch them 
> > > > > off, these holes are usually gone. But I found also images were 
> > > > > such holes appear in certain configurations, even without llyod or 
> > > > > odt active.
> > > > > There, I can influence these holes by choosing a larger 
> > > > > relative_error_bound in create_labeled_image_mesh_domain (my 
> > > > > default is 1e-7) or by generating a new mesh, as there is some 
> > > > > randomness in initialize_triangulation_from_labeled_image (in my 
> > > > > code, I specified a seed for the PRNG and thus can reliably trigger 
> > > > > it by choosing a specific seed).
> > > > >
> > > > > Interestingly, it seems to appear more often when more than two 
> > > > > labels are used. Thus, in a binary image (only 0 and 1), this does 
> > > > > not happen. For example, if I take the same image with 3 labels and 
> > > > > replace all 2's by 1's, I get a much better mesh.
> > > > >
> > > > > Is there anything I may have missed to implement, for example 
> > > > > during the creation of the label domain? I can try to track down 
> > > > > the problem even more, but don't know at the moment where to look 
> > > > > next and thus would appreciate some guidance!
> > > > >
> > > > > Best,
> > > > > Sebastian