Assistance required with Rendering error / random numbers

Carsten Bauer wrote:

Hi DAI-crew,

the artifacts stem most likely from a special random generator used for all kinds of ray jittering applied in the process of penumbra generation. I don't really know why this is done, because pictures often look rather strange with these patterns.

Basically the idea the is to generate equally distributed random numbers.
A "few" (hundred or thousand) true random numbers (generated with -DMC) are a bit "clustered", so, to cover the interval in question ( typically [0..1] ) more evenly even with a few points, an "afterburner" routine tries to equalize the distribution of these numbers. The price for this is the stupid pattern. Augustinus Topor at FhG-ISE looked into this last year too but we found no better idea yet.
-DMC avoids the pattern but the increase in necessary picture size before downsampling is substantial (easily a factor of 4). Otherwise the random clusters show up in the final image.
A somewhat dis-satisfying situation at present.

-Peter

···

--
pab-opto, Freiburg, Germany, www.pab-opto.de

......

-DMC avoids the pattern but the increase in necessary picture size
before downsampling is substantial (easily a factor of 4). Otherwise the
random clusters show up in the final image.
A somewhat dis-satisfying situation at present.

Hmmm,...

Hi Peter,

if you and Greg ascertain that there are calculational advantages (mainly
speed) in using distributed random numbers will believe it., although I
haven't noticed a considerable slowing down so far. And I cannot confirm at
all the mentioned strong effect on necessary picture size before downsampling
If the relevant parameters are set correctly, I've made very good experiences
with the -DMC version. For printing, one can even get along with
downfiltering by a factor less than two, provided the image itself is big
enough in absolute pixels. But adequate parameter setting is of course vital.

However, penumbra generation is something which affords a bit thinking in
terms of simulation, how to achieve the effect, not just putting 1:1 RealLife
to SimLife corresponding geometry into the machine and press return (apology
to Roland for using his terms without license :slight_smile: )

Additionally, when considering PCs becoming faster and faster, I wonder if
there's a reasonable way to overcome the MonteCarlo mess completey and
establish a completey different, 'deterministic' treatment of the problem
???

-Carsten

···

On Monday 05 May 2003 07:45, Peter Apian-Bennewitz wrote:

Carsten Bauer wrote:

if you and Greg ascertain that there are calculational advantages (mainly speed) in using distributed random numbers will believe it., although I haven't noticed a considerable slowing down so far. And I cannot confirm at all the mentioned strong effect on necessary picture size before downsampling If the relevant parameters are set correctly, I've made very good experiences with the -DMC version. For printing, one can even get along with downfiltering by a factor less than two, provided the image itself is big enough in absolute pixels. But adequate parameter setting is of course vital.

Well, - stratified Monte Carlo does have advantage (lower noise and faster convergence speed). "Pure" MC might be feasible per combinations of computers and projects, but I still see it's worth considering other strategies for evenly spread random numbers in Radiance.

Additionally, when considering PCs becoming faster and faster, I wonder if there's a reasonable way to overcome the MonteCarlo mess completey and establish a completey different, 'deterministic' treatment of the problem

A better, yet general, numerical integration would be appreciated by a few ...

In specular hightlights and ambient calcs the integrant is stepwise defined, non-functional with only very roughly pre-known boundaries. All I see as further enhancement is a better adaptive method. All one can do with a unknown integrant is to use already gained information as "intelligently" as possible. Educated guesses.

It might be well worth looking at the integrant more closely, though. Some statistics on the integrant found in all the sceneries of light simulations might show some patterns that suggests optimizing the integration method.
This proposal is not very scientific, but, RISC CPUs ("reduced-instruction-set") where suggested by the statistic that only a fraction of the many complex instructions of older CPUs was ever used.

-Peter

···

--
pab-opto, Freiburg, Germany, www.pab-opto.de

...

In specular hightlights and ambient calcs the integrant is stepwise
defined, non-functional with only very roughly pre-known boundaries. All

Aren't you already talking about SimLife[tm] here? In RealLife[tm] almost
everything is something smooth and integrable and discretization is
introduced deliberately to allow a numerical calculation. So your idea of
looking at the characteristics of the underlying physical processes to find a
suitable form of discretization sounds straightforward and is the way to go,
I believe.

But, oh well, just one more academic playing around with some thoughts which
won't get realized due to the usual reasons ...

-Carsten

···

On Monday 05 May 2003 15:20, Peter Apian-Bennewitz wrote:

Carsten Bauer wrote:

...

In specular hightlights and ambient calcs the integrant is stepwise
defined, non-functional with only very roughly pre-known boundaries. All
   
Aren't you already talking about SimLife[tm] here? In RealLife[tm] almost everything is something smooth and integrable and discretization is introduced deliberately to allow a numerical calculation.

From an inside viewpoint through a window, the radiance distribution is very much stepwise in reality, isn't it ?
Your argument holds if diffracting at the window border is taken in and only on a very detailed scale.

So your idea of looking at the characteristics of the underlying physical processes to find a suitable form of discretization sounds straightforward and is the way to go, I believe.

It's not so much the underlying Physics (they're quite clear), but the statistics of typical scenery.

···

On Monday 05 May 2003 15:20, Peter Apian-Bennewitz wrote:

But, oh well, just one more academic playing around with some thoughts which won't get realized due to the usual reasons ...

-Carsten

_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general

--
pab-opto, Freiburg, Germany, www.pab-opto.de

But what about the finite diameter of the sun? Ok, well, let's stop this...:slight_smile:
Those occurrences, where the lighting can be perfectly approximated by - or
more or less really is - a stepwise function normally don't cause any
problem. I solely meant diffuse and smoothly distributed lighting which is
cut into pieces and randomly thrown into the room by some MonteCarlo Chopper
...

-cb

···

On Tuesday 06 May 2003 09:38, Peter Apian-Bennewitz wrote:

From an inside viewpoint through a window, the radiance distribution
is very much stepwise in reality, isn't it ?