[Bf-cycles] Random values in SVM nodes

Brecht Van Lommel brechtvanlommel at pandora.be
Fri Mar 28 23:17:24 CET 2014


I think sampling should be left out of the shader entirely, that's the job
of the integrator. Dispersion should be implemented as a closure.

It can be useful for experimenting, but the closure based design exists to
have a clean separation here. For example the branched path tracer can take
multiple samples for the same shader evaluation, bidirectional methods
might do something similar.


On Friday, March 21, 2014, Lukas Stockner <lukas.stockner at freenet.de> wrote:

> Hi all,
> I recently noticed that a Random Value Node would be useful, for example
> for spectral dispersion shader setups.
> So now I'm thinking about implementing one, but my question is how to do
> the actual random number generation in the SVM. Using the randb value
> will probably give correlation issues, so the best approach would be to
> actually get a separate number from the RNG. However, since the number
> of random nodes used isn't known in advance, so setting aside RNG
> dimensions for these is problematic (also, allocating several dimensions
> per bounce only for the nodes is quite a waste since it reduces quality
> for the other values).
> So far, my best idea was to hash the randb value and use the resulting
> value as a seed for a LCG, however, for this every Random Node would
> have to carry an unique ID. However, the resulting samples wouldn't be
> as well-distributed as regular RNG samples...
> So, do you think the hashing idea would work, would you use another
> approach or not add a Random Node at all?
> Lukas Stockner
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20140328/a1692fd1/attachment.htm 

More information about the Bf-cycles mailing list