On Wed, 03 Apr 2002 16:45:02 +0200, Thomas Willhalm <wil### [at] fmiuni-konstanzde>
wrote:
> Patrick Elliott wrote:
> > 
> > Yeah to subsurface scattering! You can use media, but if you specifically
> > want to use a thin, single layer mesh that has no 'interior', media won't
> > work. 
> 
> Where should your subsurface scattering stop if you have only a single 
> layer? The clou with subsurface scattering is that there happens something
> to your light _below_the_surface_, right? So there has to be a notion of
> "inside" and "outside". That's why you need an object with well defined
> inside and outside to get this thing working. The "inside" modifies the
> light. That's what "media" does. So, using "media" for subsurface scattering
> is natural. (However, "media" should be extended to support other scattering
> functions.)
> 
> A "thin" mesh doesn't make much sense for subsurface scattering. In
> particular, if the mesh is infinitely thin. Then you don't have an
> "inside" at all! 
> 

Yes well... On the scale of a large object the thickness of the actual portion
of the surface that is A) visible and B) contributes to the scattering is very small,
it
makes no practical sense under such circumstances to fill the entire object with
media. This is especially true if you wanted to use a solid texture 'under' the
surface.
If the thickness of the scattering layer is 1/1000th of the objects total width it is
not practical to make it into a solid mesh with a clear interior. As it is for a
complex
object that did this you would need to place two copies of the mesh, one scaled
slightly smaller and mapped with the main texture, while the other contained the
media. Why double the memory used and use media which is much slower than
a simple simulation that can be applied to a single surface? Never mind the fact that
some shapes may make such scaling complicated or impossible. There are many
situations in which media is absolutely neccessary, but in this case it may be much
slower and more complicated than subsurface scattering really needs to be.

Not everyone that would like to use subsurface scattering can afford the hardware,
time or memory needed to do it the 'right' way. I doubt that someone on a time limit
cares if it is done right, as long as it looks right. A simpler, faster and 'close
enough'
solution would be very helpful. As it is short of coding a patch yourself there really
is
no viable alternative to doing it using media. And I suspect that many of those who
use POV, including myself, are not well equipped to create such a patch.

The arguement seems to be that it can already be done and that the existing method
does it right. I am sure that if we had fast enough computers we could also use real
simulations of tree growth, crystal growth or even molecular structures to 'correctly'
generate the patterns, IOR and color of everything in POV, but is it practical to use
such things even if they are implimented? I would say no. In this case we are asking
for an alternate, and in most cases in which it is likely to be used, far more
practical
alternative. Why is this a bad thing?

Post a reply to this message