How Pixar changed the way light works

Become a member of the CGSociety

Connect, Share, and Learn with our Large Growing CG Art Community. It's Free!

THREAD CLOSED
 
Thread Tools Search this Thread Display Modes
  06 June 2013
Originally Posted by Zendorf: Keep in mind that the site this article is on is a general tech site, not aimed at CG professionals, so the terminology has been dumbed down. I was very interested to see snippets in the accompanying video of their proprietary animation software (still called Marionette ?) as it looks much more refined than what I have seen in previous videos.

Pixar is no longer using Marionette (except in circumstances where they need to look at archived animation files). Starting with Brave, they switched over to Presto.

Originally Posted by Zendorf: Seems it is now based on viewport manipulation as compared to the old system of gui sliders. Also looks to have a very nice viewport opengl fur representation. Very nice clean looking gui with the f-curves laid out like that. Would love to know more from those in the know about this proprietary software, since it is not likely that I will ever get to use it

I'll leave it as a google exercise for you to find out what's publicly known about Pixar's Presto animation software.

Originally Posted by Zendorf: Interesting....so is it a brute force GI solution or a pre-cached irradiance map style solution?

According to the Pixar paper mentioned above by sentry66, it's a multiresolution cached radiosity solution (irradiance is just one component of this cache). They've acknowledged brute force (path tracing) GI solution used for Monster House (i.e. Arnold), but they said that approach is too noisy for their preferences.
 
  06 June 2013
Thanks for the info pomru
 
  06 June 2013
so what would be faster in 2013. pixar's new GI with displacement or a brute force optimized raytracer with displacement?
 
  06 June 2013
Originally Posted by ice-boy: so what would be faster in 2013. pixar's new GI with displacement or a brute force optimized raytracer with displacement?

I don't think they are quite comparable, and when you go in the realm of plausible shaders the shader makes a lot of difference. The more brute-force tracers also take some smart but significant shortcuts in limiting tessellation and switching to normal/bump tricks before the fragment is reached.

In general PRMan with this is basically going away from REYES and confronting itself with the problems raytracing has always needed to solve.
They make a big deal in that article of memory management and how everything has to be kept in memory, but that's exactly what detractors of raytracing were always praising REYES for (and back then, before 64GB of ram on a blade were standard it was a good point).

In terms of sheer speed I somehow doubt PRMan would be quite on par yet with someone who's been doing solely this and refining it like all hell for over a decade now, like most of the Arnold team, but sheer performance in rendering a frame isn't really the key parameter either engine wants to be judged by.

What matters is that PRMan retains its identity as developer friendly while doing that, which is no small feat, and the whole point (which to be honest Marcos, founder of SolidAngle has been championing for years) is that the time that gets shortened is iterations at the desk, which is the really expensive time, the farm time matters very little compared to surfacing and lighting team human hours.

They also went for fairly different solutions. The more brute force ones have a more predictable scaling in general and little to no precomputation, but there are plenty cases where you do need a de-noise process in the pipe, so you have a post-load to consider after the numbers have been crunched.
Pixar found it not to their taste, so they actually went for a less noisy solution which has a computational front-load instead.
__________________
Come, Join the Cult http://www.cultofrig.com - Rigging from First Principles
 
  06 June 2013
Originally Posted by pomru: , it's a multiresolution cached radiosity solution (irradiance is just one component of this cache).


Seriously?
Are we still stuck in 2003 ?
 
  06 June 2013
I think pixar's pipeline handles caching to a point where the artists don't have to "deal" with them that much, but don't quote me on that. I totally agree that arnold can put out some noisy images though. Brute forcing GI is still hella' slow in most production cases.
__________________

 
  06 June 2013
Originally Posted by staticVex: I think pixar's pipeline handles caching to a point where the artists don't have to "deal" with them that much, but don't quote me on that. I totally agree that arnold can put out some noisy images though. Brute forcing GI is still hella' slow in most production cases.

Any big PRMan pipeline I know of makes caching transparent to some extent, but regardless of that it still is a front-load, which means before things have gone through that caching process, you can't see the full results while you iterate lighting.
Conversely, post-loads like de-noising means you will usually have a slight visual change in texture and feel of things well after the iteration is done and you might need to do one more pass.

Ideally you would do neither and just brutishly compute your way to perfection. Some times you get it, some other times though it's too expensive and you have to pay either an entry or an exit fee.
Pixar simply went for the former.
__________________
Come, Join the Cult http://www.cultofrig.com - Rigging from First Principles
 
  06 June 2013
You think some time in the future they'll break down and start rendering layers?
__________________
[Invivo Animation Reel]
 
  06 June 2013
Originally Posted by ThE_JacO: Any big PRMan pipeline I know of makes caching transparent to some extent, but regardless of that it still is a front-load, which means before things have gone through that caching process, you can't see the full results while you iterate lighting.

I wouldn't put it that way:
http://graphics.pixar.com/library/R...g/globillum.m4v
 
  06 June 2013
Originally Posted by ice-boy: so what would be faster in 2013. pixar's new GI with displacement or a brute force optimized raytracer with displacement?

The whole point of caching is to be faster than brute force. It's not like path tracing was some fancy new invention, the algorithm was introduced in 1986. Methods like irradiance caching and photon mapping (both of which are based on path tracing) were all invented to overcome the slow performance of path tracing.
 
  06 June 2013
Originally Posted by stew: I wouldn't put it that way:
http://graphics.pixar.com/library/R...g/globillum.m4v


I can't speak for Pixar, as I never worked there, and I can't quite speak for Animal Logic, as (ironically enough) the place I know best is also the one I can't discuss due to NDAs In general terms though I think the front-load to post-load distinction does apply to a lot of places.
As a shot is still iterating animation or environment you do have to wait for the farm to de-que the caches before it can be moved to a lighter's desk to iterate, while with other solutions the tendency is to reach acceptable and then leave the rest to a mix of brute force and de-noise/post.

It's admittedly a bit of a generalization, and both disappear inside the schedule if line production is solid, but IMO it's a valid enough distinction between the two approaches to keep in mind, at least to the effect of not discounting what Pixar is doing as antiquated (which it's not), or to diminish the more brute force solutions as too expensive (which once desk time is factored in they often aren't).
__________________
Come, Join the Cult http://www.cultofrig.com - Rigging from First Principles
 
  06 June 2013
Originally Posted by DanHibiki: You think some time in the future they'll break down and start rendering layers?


I thought they did? My impression is that their usage of compositing is minimal, but not totally non-existent.
__________________
kevinbakercg.com

Last edited by Meloncov : 06 June 2013 at 10:50 PM.
 
  06 June 2013
Well, there is a big difference though.

At FMX, Marcos wore a T-shirt that said "Buy me a drink and you get 10% discount from Arnold"

Seen it with my own eyes
__________________
Tamas Varga
 
  06 June 2013
Sounds like him alright
__________________
Come, Join the Cult http://www.cultofrig.com - Rigging from First Principles
 
  06 June 2013
Yeah, well one of our developers deserted to him so we've been warned in advance...

Even tied to spread the word about it, and some responses were like, "damn I've just ordered a lot of licenses"

__________________
Tamas Varga
 
Thread Closed share thread



Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is On
HTML code is Off
CGSociety
Society of Digital Artists
www.cgsociety.org

Powered by vBulletin
Copyright 2000 - 2006,
Jelsoft Enterprises Ltd.
Minimize Ads
Forum Jump
Miscellaneous

All times are GMT. The time now is 01:57 PM.


Powered by vBulletin
Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.