Any thoughts on replacing the current radiosity with a new system?

If not, I would at least like the ability to up the count beyond 1600.  Although
the current maximum allows one sample approximately every 1.5 degrees (surface
area of 4*pi / 1600 samples * 180 degrees per radian), I've found that using
realistically sized objects with high ambient values to light radiosity scenes
need a finer resolution than this.

I've done some tests hacking scenes together to use only photons for lighting,
and I personally prefer this method (I should get around to posting some
samples).  Others have also mentioned BRDF, I believe it's called, where you
basically write your own lighting routine.  And I'm sure there are still other
methods for global illumination.

So, any thoughts on what should be used for POV 4?

Post a reply to this message

"Chambers" <bdc### [at] yahoocom> wrote:
> Any thoughts on replacing the current radiosity with a new system?
>
> If not, I would at least like the ability to up the count beyond 1600.  Although
> the current maximum allows one sample approximately every 1.5 degrees (surface
> area of 4*pi / 1600 samples * 180 degrees per radian), I've found that using
> realistically sized objects with high ambient values to light radiosity scenes
> need a finer resolution than this.
>
> I've done some tests hacking scenes together to use only photons for lighting,
> and I personally prefer this method (I should get around to posting some
> samples).  Others have also mentioned BRDF, I believe it's called, where you
> basically write your own lighting routine.  And I'm sure there are still other
> methods for global illumination.
>
> So, any thoughts on what should be used for POV 4?


I like the adaptive pretrace in MegaPov 1.21.  It eases the limits of having a
max count, and greatly helps in some scenes when you need a very low error
bound.  I think this or something similar would be a good way to go for
radiosity, but for povray 4, I'd like to see something else entirely.

It occurred to me the other day that Povray's current light model is very good
for some things, but starts with a number of assumptions that may not be true
for certain scenes.  Therefore, it may be better to allow the user to select a
lighting/shading model viaa the global_settings{ } block that best fits the
scene.

Some changes I would like to see for radiosity is the use of BRDF for building
non-image output data that would permit someone to re-render the scene from
different angles (assuming no other scene elements have changed) with minimal
processing.  I also like the idea of generating a 3D version of a lightmap that
could be reused for an animation (at the cost of moving objects in the scene not
being affected by and not affecting the ambient light).

There are a number of things to look at here that almost always come into play
with scenes using radiosity, especially the difficulties of simulating the wide
dynamic range of light that is often present.  Right now, with my (pretty
limited) understanding of light models, I favor the approach of having more
than one available.

-Reactor


P.S. - a semi-workaround for the 1600 count limit can be to increase the amount
of pretrace you are using, even to the point of sub-pixel pretrace.

Post a reply to this message

"Chambers" <bdc### [at] yahoocom> wrote:
> Any thoughts on replacing the current radiosity with a new system?

Of course.  That thought haunts me every time i see, "Cleanup Parse Warning:
This rendering uses the following experimental feature(s): radiosity. The
design and implementation of these features is likely to change..."

> If not, I would at least like the ability to up the count beyond 1600.  Although
> the current maximum allows one sample approximately every 1.5 degrees (surface
> area of 4*pi / 1600 samples * 180 degrees per radian), I've found that using
> realistically sized objects with high ambient values to light radiosity scenes
> need a finer resolution than this.
>
> [snip]
>
> So, any thoughts on what should be used for POV 4?

How about specifying points (or directions) near which the radiosity sampling
density would increase.  The overall count could be kept relatively low.  This
would be analogous to the photons target spacing multiplier.

Post a reply to this message