|
 |
-Hail
There are a couple of issues concerning photon mapping. You can find
source code and images in the binary section.
First issue is related to the distance between light source and target
object, when the light source moves away the photons begin to act
erratically. Each pov unit added to that distance will give completely
different results. Sometimes about half the area where the photons should
hit is a bit fuzzy and the rest is normal and other times it is fuzzy all
the way and sometimes it is correct. The resulting fuzziness is always the
same at identical settings.
The second issue has to do with the way photons are shot from area lights.
If the area_light feature is not used then the photons are shot from the
corner of the area light (first light?) and this causes caustics to fall in
obviously wrong places.
P2-450, 64Mb Ram, W2k, Pov 3.5 b-6
Post a reply to this message
|
|
|
|
"Vampyrium" <cyb### [at] hotmail com> wrote...
> First issue is related to the distance between light source and target
> object, when the light source moves away the photons begin to act
> erratically. Each pov unit added to that distance will give completely
> different results. Sometimes about half the area where the photons should
> hit is a bit fuzzy and the rest is normal and other times it is fuzzy all
> the way and sometimes it is correct. The resulting fuzziness is always the
> same at identical settings.
I'm fairly confident that this is a floating-point accuracy problem.
As distance increases, the accuracy of the ray-sphere intersection tests
decreases. The problem is that you are shooting a bunch of rays with the
objects, and the difference between all those rays is minimal... the rays
are practically parallel because the object is so small and the light source
is so far away. Floating-point accuracy problems then lead to errors in the
result.
If you scale the entire scene together, then you won't notice this problem
until you reach POV's physical limits. The issue is the distance from the
light to the spheres relative to the size of the objects.
small distance + small spheres = no problem
big distance + big objects = no problem
big distance + small objects = accuracy problems
> The second issue has to do with the way photons are shot from area
lights.
> If the area_light feature is not used then the photons are shot from the
> corner of the area light (first light?) and this causes caustics to fall
in
> obviously wrong places.
This has been fixed for the next beta.
-Nathan
Post a reply to this message
|
|
