[Bf-cycles] Antialiasing overbright areas in Cycles

[Ton Roosendaal]

Hi,

Before I added "FSA" I asked a lead ILM guy how they handle this issue, and they just render everything larger.

Basically rendering with 16 samples is almost same as render 4 x larger. The latter has benefits of allowing dynamic images to get all the composite effects before clamping it. "FSA" remains superior though, thanks to nicer sample patterns.

-Ton-

------------------------------------------------------------------------
Ton Roosendaal  Blender Foundation   ton at blender.org    www.blender.org
Blender Institute   Entrepotdok 57A  1018AD Amsterdam   The Netherlands

On 3 Mar, 2013, at 18:34, Mike Farnsworth wrote:

> You can pre-clamp pixel sample brightness before any averaging/compositing, as long as you're okay with a bias (toward dark) in the final pixel result.  Sometimes it can really do the trick for hot spec aliasing.  You'll find this problem more or less prevalent with caustic-like paths and with hot specs in motion blur.  You especially will see the problem with any pixel filters with negative lobes, but Cycles just uses Gaussian so this shouldn't be a problem.  The other thing to do is find the source of the hot sample further down the ray path, and deal with it there; often you can do that in a more energy-preserving way if it's in an indirect bounce.  I'll leave it up to your imagination what form these can take on.  Cycles already has the caustic avoidance, which is a bit of a blunt hammer, but there are similar things that can be done that keep some of the energy (and thus reduce bias).
> 
> -Mike
> 
> 
> 
> On Sun, Mar 3, 2013 at 8:11 AM, Brecht Van Lommel <brechtvanlommel at pandora.be> wrote:
> I'd like to investigate a solution here, we can in principle support
> FSA, and there's some other tricks that I'd like to try. But for now
> the only way to solve this is to handle a high resolution and scale
> down afterwards.
> 
> Brecht.
> 
> On Sun, Mar 3, 2013 at 3:24 PM, Ton Roosendaal <ton at blender.org> wrote:
> > Hi,
> >
> > There's a very practical reason why you cannot anti-alias it.
> >
> > You perceive aliasing *only* after putting it on a clamped display. If your screen would show values > 1.0 anti-aliasing would work. :)
> >
> > So the proper way to antialias is to first clamp colors of samples to display range, then merge all samples to average them. This is why Blender Internal "FSA" does a good job. So you can composite (per sample) the high dynamic range, then it clamps, and then merges the colors values.
> >
> > For Cycles this trick is not so obvious to add, since it takes 100s of samples per pixel...
> >
> > -Ton-
> >
> > ------------------------------------------------------------------------
> > Ton Roosendaal  Blender Foundation   ton at blender.org    www.blender.org
> > Blender Institute   Entrepotdok 57A  1018AD Amsterdam   The Netherlands
> >
> > On 3 Mar, 2013, at 14:40, Adriano Oliveira wrote:
> >
> >> Is there a way do better antialiasing overbright areas?
> >>
> >> PS: I don't want to clamp the highlights.
> >>
> >> Adriano A. Oliveira
> >>
> >> _______________________________________________
> >> Bf-cycles mailing list
> >> Bf-cycles at blender.org
> >> http://lists.blender.org/mailman/listinfo/bf-cycles
> >
> > _______________________________________________
> > Bf-cycles mailing list
> > Bf-cycles at blender.org
> > http://lists.blender.org/mailman/listinfo/bf-cycles
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles
> 
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles