Thanks to the ones who submitted scenes.

There are first test results of my new implemenations compared to the old 
ones:

See http://www.povworld.org/povray/aatest.html . Note that the servers 
bandwith is quite limited.

- Micha

Post a reply to this message

In article <3c73fc36@news.povray.org>, Micha Riser <mri### [at] gmxnet> 
wrote:

> I am experimenting with new antialiasing methods for povray. Now I am 
> looking for scenes where the current methods do a poor job or need too many 
> samples/pixels to produce a nice result. If you have such a scene which is 
> not too large (<100kB) and does not take hours to render please send it to 
> me (mri### [at] gmxnet) or post it to p.*.sf. I will then test if my new 
> implementations do better than the old ones.

BTW, looking through the old Ray Tracing News archives, there's plenty 
of information about this...one interesting method was to make a hash 
value from the objects intersected and shadow rays. If the hash value 
changed from one pixel to another, either the set of objects hit 
changed, or the image passed into/out of shadow. Then the colors were 
compared, and the pixel was only antialiased if the hash changed and the 
color was different. No antialiasing the tail of a black cat on a 
moonless night, but only object/shadow edges were antialiased. Sharp 
edges in the texture were left alone...these guys apparently didn't use 
textures with sharp edges. ;-)
This technique could be adapted to POV as a way of specifying different 
antialiasing parameters for object edges and textures...if the hash 
value triggers it, use one setting, but if the color difference is 
greater than a threshold (and the hash values are the same), use a 
different setting. If it could be further controlled...certain textures 
or objects using no antialiasing...it could make things like starfields 
much easier.

This reminds me...are the colors clipped before or after antialiasing? 
It should be after...a "superbright" object (rgb > 1) should contribute 
more to a pixel than a dim (rgb 1) one.

-- 
Christopher James Huff <chr### [at] maccom>
POV-Ray TAG e-mail: chr### [at] tagpovrayorg
TAG web site: http://tag.povray.org/

Post a reply to this message

> In article <3c73fc36@news.povray.org>, Micha Riser <mri### [at] gmxnet>
> wrote:
> This reminds me...are the colors clipped before or after antialiasing?
> It should be after...a "superbright" object (rgb > 1) should contribute
> more to a pixel than a dim (rgb 1) one.

IMHO it would be a _MAJOR_ design-flaw, if it wouldn't be this way ...

Post a reply to this message

> See http://www.povworld.org/povray/aatest.html . Note that the servers
> bandwith is quite limited.

So what can we conclude ?

1. These Scenes _are_ hard nuts for AA
2. Your method _CAN_ improve the visual quality by taking not to much  
samples ... (under circumstances)
2'. But it can also be that it makes the result even worse ...
2". The result depends on the choosen AA-Method; There's no "Best-Of" ...
3. POV Raytracing works in a diskreet world where it's hard to simulate  
the continueity of the pure math ...
... leading to:
4. proving (of course, not a mathiematical proove) right, with my former  
statement, that every known AA-Algo can be broken with a scene ...

... at least if you consider the resolution to be independend ...


BUT: Good work ... I think we needed something like that ...

Post a reply to this message

Christopher James Huff wrote:

> BTW, looking through the old Ray Tracing News archives, there's plenty
> of information about this...one interesting method was to make a hash
> value from the objects intersected and shadow rays. If the hash value
> changed from one pixel to another, either the set of objects hit
> changed, or the image passed into/out of shadow. Then the colors were
> compared, and the pixel was only antialiased if the hash changed and the
> color was different. No antialiasing the tail of a black cat on a
> moonless night, but only object/shadow edges were antialiased. Sharp
> edges in the texture were left alone...these guys apparently didn't use
> textures with sharp edges. ;-)

I do not like the idea of antialiasing only at objects frontiers.. as 
textures are certainly one of the most important thing with POV-Ray.

> This technique could be adapted to POV as a way of specifying different
> antialiasing parameters for object edges and textures...if the hash
> value triggers it, use one setting, but if the color difference is
> greater than a threshold (and the hash values are the same), use a
> different setting. If it could be further controlled...certain textures
> or objects using no antialiasing...it could make things like starfields
> much easier.

I do not know the povray source good enough for adding object specific 
features :(.

> This reminds me...are the colors clipped before or after antialiasing?
> It should be after...a "superbright" object (rgb > 1) should contribute
> more to a pixel than a dim (rgb 1) one.

I will look this up.

- Micha

Post a reply to this message

Jan Walzer wrote:

>> See http://www.povworld.org/povray/aatest.html . Note that the servers
>> bandwith is quite limited.
> 
> So what can we conclude ?
> 
> 1. These Scenes _are_ hard nuts for AA

Sure.

> 2. Your method _CAN_ improve the visual quality by taking not to much
> samples ... (under circumstances)
> 2'. But it can also be that it makes the result even worse ...

Well, my method is based on method 2, that is also adaptive. If you compare 
it with the standart method 2, my new implementations do always equally or 
better. (sure one could contruct a scene where this is not the case)

> 2". The result depends on the choosen AA-Method; There's no "Best-Of" ...

I have been a bit surprised how god method 1 actually performed.. as it is 
really simple.

> 3. POV Raytracing works in a diskreet world where it's hard to simulate
> the continueity of the pure math ...
> ... leading to:
> 4. proving (of course, not a mathiematical proove) right, with my former
> statement, that every known AA-Algo can be broken with a scene ...
> 
> ... at least if you consider the resolution to be independend ...

This is certainly true.. but the worst case scenario is here certainly not 
allowed to be the criterion.

> 
> 
> BUT: Good work ... I think we needed something like that ...

Thanks.

Post a reply to this message

"Christopher James Huff" wrote:
> This reminds me...are the colors clipped before or
> after antialiasing?
> It should be after...a "superbright" object (rgb > 1)
> should contribute more to a pixel than a dim (rgb 1) one.

It is clipped before AA and there's a good reason.

Suppose you have a really bright sphere (the sun). All pixels that hit the
sun would be completely white no matter if the sun occupies a small or big
amount of the pixel, simply because the high brightness would dominate in
the average. The result would be that very bright objects would not look
antialiased at all no matter how high quality AA settings you used.

Rune
--
3D images and anims, include files, tutorials and more:
Rune's World:    http://rsj.mobilixnet.dk (updated Feb 16)
POV-Ray Users:   http://rsj.mobilixnet.dk/povrayusers/
POV-Ray Webring: http://webring.povray.co.uk

Post a reply to this message

I'll be looking at this more closely tommorow. What little I see is very
promising. Thanks for taking my proposal seriously Micha! This should be
perfect for the next MegaPOV. :-D

Post a reply to this message

In article <3c782d3c@news.povray.org>,
 "Rune" <run### [at] mobilixnetdk> wrote:

> It is clipped before AA and there's a good reason.
> 
> Suppose you have a really bright sphere (the sun). All pixels that hit the
> sun would be completely white no matter if the sun occupies a small or big
> amount of the pixel, simply because the high brightness would dominate in
> the average. The result would be that very bright objects would not look
> antialiased at all no matter how high quality AA settings you used.

Well, this explains why I never succeeded in getting stars to survive AA 
by increasing their brightness...that was the exact effect I expected 
and wanted. I can manually soften the sun's outline (media, or a dimmer 
sphere visible just around its edges), eliminating the problem there, 
but there doesn't seem to be any way to compensate for small 
(sub-pixel), very bright objects getting obliterated by the antialiasing.
Maybe a smarter algorithm would be possible...if there are other 
high-brightness pixels nearby, you are at the edge of a bright object, 
so clip and supersample, otherwise supersample then clip.

-- 
Christopher James Huff <chr### [at] maccom>
POV-Ray TAG e-mail: chr### [at] tagpovrayorg
TAG web site: http://tag.povray.org/

Post a reply to this message

In article <8JP### [at] lzernet>, jan### [at] lzernet (Jan Walzer) wrote:

> 4. proving (of course, not a mathiematical proove) right, with my former  
> statement, that every known AA-Algo can be broken with a scene ...

Just use an infinite number of samples per pixel. I don't have the CPU 
power to test this though, ask God. ;-)
Actually, that might not be enough, because you still have the grid of 
pixels...how about creating a neural net to detect the moire patterns 
and compensate for them?
Or just reduce the pixel size to near-microscopic. Then you won't even 
need antialiasing. ;-)
Really, I think resolutions are going to make a jump soon, so 
antialiasing won't be as big a requirement.

-- 
Christopher James Huff <chr### [at] maccom>
POV-Ray TAG e-mail: chr### [at] tagpovrayorg
TAG web site: http://tag.povray.org/

Post a reply to this message