Hello,

Please find a picture to illustrate my question (this is not my submission
for povcomp :-))
-> http://site.voila.fr/jcolin/galerie/GI/haltonrandom.jpg

and the source file for this picture (I can hardly make it simplier)
-> http://site.voila.fr/jcolin/Fichiers/cuisine.zip


The scene is a sort of closed box with a open window which illuminates the
room. No usual light source.
Used Feature: Radiosity with Randomized radiosity sampling directions (with
Halton sequence). The radiosity settings seem weird but it is only intended
to highlight that used feature.

Actually, I do not understand why the vertical wall (mesh) seems to show the
effect of random rotation of sample ray direction (which leads to the
expected noisy looking illumination) whereas the plane primitive for ground
has a regular "projection-like" illumination.

Kindly look at the following picture for comparison(It is the same scene
with Randomized radiosity sampling directions feature disabled):
-> http://site.voila.fr/jcolin/galerie/GI/haltonnonrandom.jpg
The ground/plane illumination is not changed.

My computer and my POV version :
Megapov 1.1
WindowsXP
512 Mo Ram
Intel 2400 MHz

By the way, and for this case, the same scene rendered with the internal
sample directions does not seem to give an uniform distibution of the 50
samples. which has already been reported in
http://www.8ung.at/halcyon/povray/custom-radiosity.html.
-> http://site.voila.fr/jcolin/galerie/GI/internalrandom.jpg

In addition, I noticed the recent issue raised by George Pantazopoulos too
(randomized sample direction rotation which is less effective for samples
taken from a near-normal direction. The effect is noticeable in scenes with
small bright sources (in a "cornell box"-like scene for instance))


Best Regards,

Jerome

Post a reply to this message

>
> Actually, I do not understand why the vertical wall (mesh) seems to show the
> effect of random rotation of sample ray direction (which leads to the
> expected noisy looking illumination) whereas the plane primitive for ground
> has a regular "projection-like" illumination.
>

Jerome,

Excellent work (and timing)! The above effect you illustrate is exactly the
effect I was talking about in my recent post. I'll bet the normals on the
ground are all facing in the Z direction. The Megapov radiosity code does
not do the random rotation if the surface normal is pointing near the Z
axis. Since your ground is flat, all the normals point the same way. That
is, they are all parallel to the world's Z axis. Thus none of them get the
random rotation that you requested in your radiosity {} block. I believe
this is a bug in the code and your picture has helped convince me. Thanks!


> In addition, I noticed the recent issue raised by George Pantazopoulos too
> (randomized sample direction rotation which is less effective for samples
> taken from a near-normal direction. The effect is noticeable in scenes with
> small bright sources (in a "cornell box"-like scene for instance))
>
>
That isn't the effect I described in my recent post. The problem I described
is that if the surface point's normal happens to be (nearly) parallel to
the World Z axis, no random rotation is done at all. I'd be interested in a
picture and more info about this particular effect.

Regards,
George

http://www.gammaburst.net

Post a reply to this message