>On 17 Jun 1998 01:25:26 +0200, bob### [at] casimirrezelenstfr (Roland Mas)
>wrote:
>
>>Roy Schulz <roy### [at] rcsurztu-dresdende> writes:
>>> 2. Better radiosity with multiple interreflections. At
>>> http://www.lightscape.com are nice samples how it should look. (This is
>>> no critics. I know that radiosity is only experimental in POV 3.0)
>
>Stephen Horn claims to be working on a better radiosity model.  Until
>then, I would like to see the bug that causes it to turn off radiosity
>for all future renders when you cancel a render in POVWin fixed.
>

Many are already out there!   Search for "A Ray Tracing Solution for Diffuse
Interreflection"  by  Gregory J. Ward,  Francis Rubenstein,  and Robert D.
Clear.   This is work and discussion from Siggraph '88   This is the best
argument I've seen for telling _which pixels_ should have a radiosity
calculation performed on them.   

For a fuller and more up-to-date theoretical discussion of the problem search
for the Ph.d dissertation by Eric P. Lafortune, titled "Mathematical Models
and Monte Carlo Algorithms for Physically Based Rendering"   Make sure to have
a nice Post-script file viewer.  (I suggest GhostView and GhostScript for you
WIN95'ers)

My major complaint against Monte Carlo is that it defines second-level points
to contain contributions from only one direction.  Second-level points should
contain an interpolated contribution from ALL first-level points.  There is a
similar problem with first-level points; they contain contribution from only a
piece of the light sources, not a full interpolation of the myriad of the
light sources (be that the case. Consider a scene lit by a green and a red
light source...how does Monte Carlo react?).     Also, Lafotune's
bi-directional algorithm considers a point at a time, not at all considering
the _total amount_ of rays traced for the rendering of the entire scene.
Obviously, the slowness of the radiosity calcs will be apparent to a user from
the total amount of rendering time. 

My solution to the total radiosity calculation time are databases stored as a
BSP trees.  The problem of 'best contributers' still occupies me.  More
later... 

Steve Horn

Post a reply to this message