"SharkD" <nomail@nomail> wrote in message 
news:web.487abbde1771580b302c26d00@news.povray.org...
>
> It looks incredible! I'm wondering what the results would be in terms of
> rendering times if you used a mesh instead of an isosurface. I know 
> there's an
> external utitlity to replace parametric objects with meshes. I can't 
> recall if
> there's also one for isosurfaces.
>
> Of course, the typical parametric curve tends to be much smoother than the
> object you've created here. Not sure if this is the ideal application for
> meshes.

Well theoretically it would be quicker, it depends how much detail you need. 
If you want triangles that are about 2 pixels big in the finished image then 
it probably isn't going to be much quicker, since you'd have to create 
triangles for the back and lots of areas that are out of sight (because you 
don't know what angle the mesh will be viewed from), so you'd still end up 
sampling the shape about as many times as the isosurface does...

Though actually I'm probably way off with that... I was forgetting about all 
the extra rays it traces for shadows and radiosity. e.g. Assuming you go 
with a volumetric approach, something like a marching cubes algorithm, I 
think that will sample the pattern much more efficiently since each point in 
the grid of samples would only be accessed once... But then the question is 
how slow is it to render a mesh of that complexity? And how much detail can 
we use before pov runs out of memory? And more to the point, do I really 
need it to render quickly when I have a quad processor machine at work 
that's got nothing to do at the weekend? ;)

Of course, the heightfield I rendered is technically a mesh made from 
sampling it, but that's not volumetric so the little round holes get 
stretched into deep pits. It looks similar, and is a hell of a lot faster, 
but it's not quite right.

-- 
Tek
http://evilsuperbrain.com

Post a reply to this message