I think this is really difficult. The only example I know
is the Cornell box: you try to reproduce a very basic scene,
(esay to model and reproduce exactly, with simple texture (only color))
and you take the photo and the 3d image, you subtract them:
if the result is perfect you get a really black image if not,
you can measure the proportion of non-black pixel and the
gap between each non-black pixel and the black color.

But to do this with "real" scene, this is not easy.
Model object can be done really precisely (one can
even use 3d scanner). But put texture exactly as there
are on the object, this is not easy even if you use image
map from this object. But this is feasible I think.
But I don't know any example.

For commercial software, I suppose this is less "scientific".
But for them, rapidity of rendering and the feeling of reality
(which is not reality) are really important. Moreover there
are also the modeler which play a crucial role in the development.

For now one of the best renderer for commercial product is Mentalray
(http://www.mental.com). It is used in Softimage and now in 3DS Max.
With it you have photon, radiosity, global illumination and 
the time for rendering are rather good.

Fabian.


> 
> Fabian BRAU wrote:
> 
> >This image just
> >show the quality of the renderer not the modeler.
> 
> I always wonder, so I ask it here, what are objective (measurable?)
> criteria to determine the quality of a renderer/tracer/..
> 
> Ingo
> 
> --
> Photography: http://members.home.nl/ingoogni/
> Pov-Ray    : http://members.home.nl/seed7/

Post a reply to this message