"Christoph Hormann" <chr### [at] gmxde> schrieb im Newsbeitrag
news:4126458a$1@news.povray.org...
> Thies Heidecke wrote:
> > Hi,
> > as announced in the MegaPOV-Thread i've made a small webpage
> > to show my new radiosity-sampleset. The URL is:
> >
> >   http://www.8ung.at/halcyon/povray/custom-radiosity.html
> >
> > What do you think?
> > I'm excited to get your feedback :)
>
> That's a regular distribution and this means it is a very bad one for
> general purpose.  This does not mean it can't lead to better results
> under certain circumstances.
>
> In case it isn't immediately clear why a regular distribution is bad -
> the lower edge of your sample set for example follows a certain pattern
> - this pattern is the most likely cause of artefacts.

Although one can see the typical spiralling patterns of phyllotaxis it isn't
(considered on a certain level) as regular as it seems. In fact, the pattern
is, because of the golden ratio involved, angle-wise the most irregular

sampling
don't exist with the sunflower-pattern.

Could you show me an example-scene where these kind of artifacts caused by
regular-distribution sampling can be seen?

> Christoph Hormann
Greetings,
Thies

Post a reply to this message

Thies Heidecke wrote:
> Hi,
> as announced in the MegaPOV-Thread i've made a small webpage
> to show my new radiosity-sampleset. The URL is:
> 
>   http://www.8ung.at/halcyon/povray/custom-radiosity.html
> 
> What do you think?
> I'm excited to get your feedback :)
> 
> Greetings,
> Thies
> 
> 

This is very interesting, Thies! Could you also test with an indoor 
scene? How about a room with a small window letting in bright sunlight?

Regards,

-- 
-------------------------
George Pantazopoulos
http://www.gammaburst.net

Post a reply to this message

"George Pantazopoulos" <geo### [at] seemysignature> schrieb im Newsbeitrag
news:41275df6$1@news.povray.org...
> Thies Heidecke wrote:
> > Hi,
> > as announced in the MegaPOV-Thread i've made a small webpage
> > to show my new radiosity-sampleset. The URL is:
> >
> >   http://www.8ung.at/halcyon/povray/custom-radiosity.html
> >
> > What do you think?
> > I'm excited to get your feedback :)
> >
>
> This is very interesting, Thies! Could you also test with an indoor
> scene? How about a room with a small window letting in bright sunlight?

Hi,
thanks for your interest!
A third scene is underway at the moment. I'm currently rendering the
reference-image with count 6400, that will take a few hours, then i'll
make a series of images to compare like in the other scenes.
The scene consists of a simple closed room which is divided by a row
of columns and a strong lightsource on one side of the room and the
camera on the other side.
I'm also considering making images with the halton-sequence-set to
compare. I'll let you know when i've uploaded the new scene.

> Regards,
> George Pantazopoulos

Greetings,
Thies

Post a reply to this message

I actually tried this golden ratio approach two years ago iirc and our brain 
is very good at detecting patterns and regular artifacts, so unfortunately 
it doesn't work so good for low sample amounts. But for very high sample 
counts I think it's ok though.

Post a reply to this message

"Eli" <eli### [at] jehoelnet> schrieb im Newsbeitrag
news:41276939@news.povray.org...
> I actually tried this golden ratio approach two years ago iirc and our
brain
> is very good at detecting patterns and regular artifacts, so unfortunately
> it doesn't work so good for low sample amounts. But for very high sample
> counts I think it's ok though.

Perhaps you and christoph are right about the regular-pattern-thing. The
reason why i'm so sceptical until now is, that i haven't seen a scene yet
where my approach looks worse than the internal sampleset. Although the
reason for that could be that my scenes don't show that artifacts off
good enough (i think the indoor-scene will give more possibilities to
compare). But i must say that especially at low count-values i found the
golden-ratio approach not looking very regular.

Greetings,
Thies

Post a reply to this message

The Cornell box is a good test scene.

I think a sample set should work at low sample counts.

Maybe jittering the set in order to get rid of the visible patterns should 
help.

Post a reply to this message

"Eli" <eli### [at] jehoelnet> schrieb im Newsbeitrag
news:4128b0aa$1@news.povray.org...
> The Cornell box is a good test scene.
I will first finish the renderings based on my current indoor-scene
because i've already rendered a few images and hours.

> I think a sample set should work at low sample counts.
What do you mean? I don't understand.
If you refer to my last sentence in the last post, i wanted
to say that especially at low sample-numbers it is hard to
detect a pattern in the golden-ratio approach so i think it's
quite good at low sample-values

> Maybe jittering the set in order to get rid of the visible patterns should
> help.
That's a good idea which i'm already trying at the moment. Perhaps this
is the most flexible solution. Combining the good distribution of the
goldenratio-method with the natural artifact-avoiding look of random
placement of samples.

Greetings,
Thies

Post a reply to this message

I've updated my webpage,
the indoor-scene is now ready to be watched.
I made it fullsize this time to see the artifacts
more clearly. The size of all images for scene 3
is about 3 MB.

I don't know if this scene is sufficient to
compare certain artifacts, if you have suggestions
for other scenes or have comments on this one
i would like to hear them !

Thies

Post a reply to this message

Thies Heidecke wrote:
> I've updated my webpage,
> the indoor-scene is now ready to be watched.
> I made it fullsize this time to see the artifacts
> more clearly. The size of all images for scene 3
> is about 3 MB.
> 
> I don't know if this scene is sufficient to
> compare certain artifacts, if you have suggestions
> for other scenes or have comments on this one
> i would like to hear them !

It looks quite good.  Note however that for a 'good looking' results it 
is less relevant how correct the results really are (i.e. how strong the 
difference to the precise solution is) and more important if there are 
clear artefacts.  While in general your sample set seems to lead to less 
differences the artefacts at low counts are significant.  Have you 
tested it in combination with the random rotation ('randomize on')?  How 
about the internal low discrepancy (halton) distribution?

Christoph

-- 
POV-Ray tutorials, include files, Sim-POV,
HCR-Edit and more: http://www.tu-bs.de/~y0013390/
Last updated 06 Jul. 2004 _____./\/^>_*_<^\/\.______

Post a reply to this message

"Christoph Hormann" <chr### [at] gmxde> schrieb im Newsbeitrag
news:ch9bij$2tl$1@chho.imagico.de...
> It looks quite good.
Thanks!

> Note however that for a 'good looking' results it is less relevant how
> correct the results really are (i.e. how strong the difference to the
> precise solution is) and more important if there are clear artefacts.
Yes, in general i agree with you, artefacts should be avoided, even if
it's not looking optimal from a mathematical point of view, since our
perception isn't mathematical either. But i think it's also important
to get an early impression of the final look without having to use
high count-values.
I also think that in most cases avoiding artefacts and good lighting
don't have to contradict each other.

> While in general your sample set seems to lead to less
> differences the artefacts at low counts are significant.
hm, i don't know exactly which images/artefacts you mean.
I think they aren't more significant than the builtin-samples-artefacts
except for the bright white stripes, which i think are pure luck/badluck.
They could have been in the internal set, too.

> Have you tested it in combination with the random rotation
> ('randomize on')?  How about the internal low discrepancy
> (halton) distribution?
That's a good idea, i totally missed that!
I tested a bit with the random rotation and it looks nice.
It makes for a totally different look, nice.
I've made a few tests with the halton-sequence, too. I think
it lies somewhere between the internal set and the golden section set.
I also experimented with jittered versions of the sunflower-set
but there are too many possible combinations =)
Seriously, perhaps i'll make another webpage to compare the various
possibilities, but probably not as detailed as the other 3 scenes, but
just a few images for each comparison.

> Christoph
Thies

Post a reply to this message