[Radiance-general] Re: light maps in radiance
Peter Apian-Bennewitz
[email protected]
Sat, 23 Mar 2002 21:43:55 +0100
--------------BBEF16751F2619BE44B30EAF
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Carsten Bauer wrote:
> Hi Peter !
>
> you're right but let's spend my last cent on this -- not if the fourth
> parameter is -1 ?
oops.- right. (never used it myself, so.)
> There is of course a personal interest behind it. The reason, why I'm
> hacking on this is, that I spent some time thinking of other ways to
> deal with broad area sources (for some design-case studies) . If you use
> glow sources in direct mode, you will see the effects of
> source-subdivision on nearby objects illuminated by them (e.g. semi-
> transparent layers. ) If you use glows in ambient mode, things are
> better, but for a nice look you need very high parameter settings,
> (although personally I'm still not satisfied ..). I got the assumption
> that the hemisphere-sampling method works better in the "far field
> range" than for near field illumination. (sorry, but it seems that I
> always try to use tools for tasks they weren't intentionally made for...
What do you mean by "far field" ? -
Quick thoughts from the back of my head: The Monte Carlo integration in
ambient calcs cause noise when the integrand in the integral over the
hemisphere of incident radiance (times BRTF) fluctuates a lot. Typically
this is a problem in indirect illumination scenes with bright spots of
direct illumination on ceilings etc.
For large area sources whose radiance is nearly constant over its surface,
the MC integration should work well. Unless one maps a brightfunc/brightpic
onto the large area undulating radiance output dependent on surface
position and thereby causing fluctuations in the integrand of nearby
ambient calcs which mean noisy ambient values there.
The MC integration is brute force (e.g. it doesn't reuse any information
from one integral to the next [would be tricky anyway], there's no real
adaptive subdivison, and with varying area sources it "under-samples" the
source at points leading to alias [maybe level-of-detail handing in
colorpicts would help there]), plus MC is slow in general, with the
positive aspects being that it's simple (which is very helpful during
validation), general and bias free. Greg's razor to Radiance had been that
it's gotta be simple and effective.
What were your conclusions and suggestions on this ?
-Peter
--
pab-opto, Freiburg, Germany, www.pab-opto.de
--------------BBEF16751F2619BE44B30EAF
Content-Type: text/html; charset=us-ascii
Content-Transfer-Encoding: 7bit
<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
Carsten Bauer wrote:
<blockquote TYPE=CITE>Hi Peter !
<p>you're right but let's spend my last cent on this -- not if the fourth
<br>parameter is -1 ?</blockquote>
oops.- right. (never used it myself, so.)
<blockquote TYPE=CITE>There is of course a personal interest behind it.
The reason, why I'm
<br>hacking on this is, that I spent some time thinking of other ways to
<br>deal with broad area sources (for some design-case studies) . If you
use
<br>glow sources in direct mode, you will see the effects of
<br>source-subdivision on nearby objects illuminated by them (e.g. semi-
<br>transparent layers. ) If you use glows in ambient mode, things are
<br>better, but for a nice look you need very high parameter settings,
<br>(although personally I'm still not satisfied ..). I got the assumption
<br>that the hemisphere-sampling method works better in the "far field
<br>range" than for near field illumination. (sorry, but it seems that
I
<br>always try to use tools for tasks they weren't intentionally made for...</blockquote>
What do you mean by "far field" ? -
<br>Quick thoughts from the back of my head: The Monte Carlo integration
in ambient calcs cause noise when the integrand in the integral over the
hemisphere of incident radiance (times BRTF) fluctuates a lot. Typically
this is a problem in indirect illumination scenes with bright spots of
direct illumination on ceilings etc.
<br>For large area sources whose radiance is nearly constant over its surface,
the MC integration should work well. Unless one maps a brightfunc/brightpic
onto the large area undulating radiance output dependent on surface position
and thereby causing fluctuations in the integrand of nearby ambient calcs
which mean noisy ambient values there.
<br>The MC integration is brute force (e.g. it doesn't reuse any information
from one integral to the next [would be tricky anyway], there's no real
adaptive subdivison, and with varying area sources it "under-samples" the
source at points leading to alias [maybe level-of-detail handing in colorpicts
would help there]), plus MC is slow in general, with the positive
aspects being that it's simple (which is very helpful during validation),
general and bias free. Greg's razor to Radiance had been that it's gotta
be simple and effective.
<br>What were your conclusions and suggestions on this ?
<p>-Peter
<pre>--
pab-opto, Freiburg, Germany, www.pab-opto.de</pre>
</html>
--------------BBEF16751F2619BE44B30EAF--