Unified sampling in depth:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.153.7536
<http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.153.7536&rep=rep1&;
type=pdf> &rep=rep1&type=pdf

 

 

From: softimage-boun...@listproc.autodesk.com
[mailto:softimage-boun...@listproc.autodesk.com] On Behalf Of Ed Manning
Sent: Tuesday, February 26, 2013 4:54
To: softimage@listproc.autodesk.com
Subject: Re: Octane render

 

I thought exactly the same, because of Autodesk's idiotic documentation.  

It is a fundamentally different approach to sampling for MR.  As I
understand it (smart people, plese help me out here!), MR formerly simply
added or in some cases, multiplied the number of samples cast from each ray
hit.  Say you have a raytraced shadow-casting area light with 3x3 samples,
and a glossy surface with sampling set to 16, and adaptive AA max at 2
(which is 4x4=16 samples per pixel) -- for a given shading fragment
(basically the intersection of a ray with a triangle), in a single bounce,
you could end up with something like ((3*3)+(3*3))*16*16 = 4608 samples
being needed just to return a value for a single ray. I'm sure I'm
oversimplifying and probably have the math wrong to boot, but basically,
every time you added something that required multiple samples -- glossiness,
motion blur, DOF, AO, soft shadows, bump maps -- you ended up increasing the
number of samples MR would take at any given point, regardless of whether
they were neededto achieve a "correct" result.

Unified sampling essentially looks at all the samples being called for by
every shader that affects a ray intersection, and somehow (math again!)
balances (and perhaps shares?) the sample values returned as they
accumulate, so that only as many samples as the user sets as a maximum, or
fewer, if the quality (think of it as 1/noise) threshold is reached, are
ever taken.  

Or you can think of it as magic.  But it can result in HUGE improvements in
render time. Sometimes.

 

 

On Mon, Feb 25, 2013 at 10:33 PM, Steven Caron <car...@gmail.com> wrote:

i stopped using mental ray when this came out, i just assumed it meant
unification of the different rendering engines. is that so or is it
something else?

 

 

On Mon, Feb 25, 2013 at 7:21 PM, Jason S <jasonsta...@gmail.com> wrote:


I  heard Sony Imagework's version of Arnold does have a unified Sampling
equivalent.

 

Reply via email to