"clipka" <nomail@nomail> wrote:
> > > Did you try "normal on" in the radiosity block? That *should* do it.
> >
> > Two other users give me that info too, thanks again. I've enabled it,
> > without a direct success. Additionally I've increased bump_size by a
> > factor of 4 and I see normals now. But it seems, the render time is
> > increased by the same factor :-)

Makes me think, there are some drawbacks with how POV-Ray radiosity currently
goes about with normals:

- If the bumps are small compared to radiosity sample density, they tend to be
blurred out.

- The effect may be non-uniform, giving stronger normal effects in corners than
on flat surfaces (due to non-uniformity of sample density).

- There's an implementation problem with radiosity and normals when using
multi-layered textures with different normal pertubations.

- Although theoretically all that normal perturbations do is affect the
weighting of the light coming in from different directions, POV-Ray just
braindeadly takes more samples (to cover different normal angles), shooting new
sample rays for each one, instead of somehow re-using the information already
gathered and just adapting it to the different (fake) surface orientation.

There must be *some* smart way to overcome all these limitations at the same
time.

Something like storing for each radiosity sample not only the resulting
illumination for a perfectly unpertubed normal, but for an additional set of,
say, a dozen standardized pertubed-normal directions, and interpolating among
these when computing the radiosity-based illumination for a particular point
that has a pertubed normal.

Post a reply to this message