[Bf-cycles] Interactive GI filtering paper

[Brecht Van Lommel]

There's a lot of such filtering papers and it's closely related to
irradiance caching. Certainly such methods would be useful for faster
previews and rendering still images. There's simple algorithms and
complex ones with different tradeoffs and inaccuracies. If people want
to contribute code for this kind of thing that can fit fine.

For animations however any method that does such filtering is tricky
to get working without flickering / low frequency noise. The new
papers in this area do not actually improve accuracy / reliability
over irradiance caching as far as I can tell, they're worse but more
general or quicker. Irradiance caching with all the tricks that have
been developed for it over the years might still be the least flickery
of all for the problem of diffuse indirect light.

I don't know of other reliable algorithms for animation that also
provide a big speedup for the simple cases. Mostly it seems that
Cycles could handle the difficult geometry, shader lighting cases
better with various changes, but if you want a big speedup for the
simpler cases then some sort of filtering or caching method with the
associated tradeoffs is the only solution I know of.

Brecht.




On Tue, May 6, 2014 at 10:35 PM, Marco G <digitalbath86 at hotmail.com> wrote:
> Hi all,
> just stumbled upon this paper, i'd be curious to know what Cycles devs think
> about such technique,
> results seems quite accurate and unlike other filtering/reconstruction
> methods, it is done online interactively.
>
> link - http://graphics.berkeley.edu/papers/Udaymehta-IPB-2013-07/index.html
>
> Would something like this be unusable, introducing a lot of inaccuracy? (or
> complexity to the code..?)
> Speaking more in general, isn't there any idea (for the future) to
> help/simplify diffuse-indirect calculation? (not Irradiance cache related,
> but something animation-friendly)
>
> Thanks,
> regards.
>
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles
>