PDA

View Full Version : Hal Bertram demonstrating "Interactive Renderman" (this is crazy)


fr3drik
08-07-2005, 03:20 PM
Exactly what is this?! Renderman taking advantage of a full renderfarm, in interactive mode? Can anyone please shed some light on this?

Original link:
http://www.deathfall.com/article.php?sid=5618

Quicktimes:
Sampling (http://www.deathfall.com/download/video/IRM/IRMsampling.mov)
Light blur and occlusion (http://www.deathfall.com/download/video/IRM/IRMlightBlurAndOcclusion.mov)
Camera interaction (http://www.deathfall.com/download/video/IRM/IRMcamera.mov)
Depth of field (http://www.deathfall.com/download/video/IRM/IRMdof.mov)
Material interaction (http://www.deathfall.com/download/video/IRM/IRMmaterial.mov)
Renderfarm tank (http://www.deathfall.com/download/video/IRM/IRMrenderfarmTank.mov)
Light interaction (http://www.deathfall.com/download/video/IRM/IRMlighting.mov)
Renderfarm volumetric light (http://www.deathfall.com/download/video/IRM/IRMrenderfarmVolumetrics.mov)

rebo
08-07-2005, 03:54 PM
Sorry Gelato you did your best but the now the daddy is showing how its done.

ThomasMahler
08-07-2005, 05:53 PM
This material interaction movie is damn impressive.

That's how a renderer should work - I do all the creative stuff, hit a button and boom, got my results. Very neato. The only question left is... what did we see here?

fr3drik
08-07-2005, 06:03 PM
At the end of one of the quicktime movies, it says "code, shaders and movies will appear at http://www.halbertram.com/interactive/".

...still nothing there.

Balusilustalu
08-07-2005, 06:08 PM
It's PRMan meets FPrime. :) New rendering technologies are coming thick and fast these days. What are they running it on though...?

Bonedaddy
08-07-2005, 06:46 PM
I was there, he explained a bit of how it's done... basically, he had (I believe) a lens shader with a point right in front of the camera on a while() loop that never ended -- so the render keeps going no matter what. Then all the standard renderman attributes used custom functions he'd written to get position, camera frustrum, etc. This kept them constantly updating. He had sliders that he could move around to get it to update in the viewport. It updated in buckets, which meant he could send it out to a renderfarm, although it was far from optimized, and resulted in immense network slowdown. Most of the time, he did it on a single workstation, all on the CPU.

I'm not the most experienced renderman guy, but he said the entire thing was being raytraced, so I believe that the point he was talking about in front of the camera is basically a transparent plane that lays on top of the camera lens.

This isn't as fast as a lot of the hardware renderers can be, and it's far from optimized. The renderer redoes the entire image, even if you're only tweaking the shaders on one object. I'm sure there will be refinements of this technique down the road, but it's still pretty damn cool. With Renderman 13, there's even the possibility to toss some shading functions to the GPU, and I'm sure once people figure out what works and what doesn't, this technique is going to beat feet.

Hats off to Hal!

Hazdaz
08-07-2005, 07:21 PM
Very very interesting. That is what rendering should be like.

FloydBishop
08-07-2005, 10:17 PM
That's very impressive. How long until we see this kind of thing incorporated into games?

rendermaniac
08-07-2005, 10:49 PM
I was there, he explained a bit of how it's done... basically, he had (I believe) a lens shader with a point right in front of the camera on a while() loop that never ended -- so the render keeps going no matter what. Then all the standard renderman attributes used custom functions he'd written to get position, camera frustrum, etc. This kept them constantly updating. He had sliders that he could move around to get it to update in the viewport. It updated in buckets, which meant he could send it out to a renderfarm, although it was far from optimized, and resulted in immense network slowdown. Most of the time, he did it on a single workstation, all on the CPU.

I'm not the most experienced renderman guy, but he said the entire thing was being raytraced, so I believe that the point he was talking about in front of the camera is basically a transparent plane that lays on top of the camera lens.

This isn't as fast as a lot of the hardware renderers can be, and it's far from optimized. The renderer redoes the entire image, even if you're only tweaking the shaders on one object. I'm sure there will be refinements of this technique down the road, but it's still pretty damn cool. With Renderman 13, there's even the possibility to toss some shading functions to the GPU, and I'm sure once people figure out what works and what doesn't, this technique is going to beat feet.

Hats off to Hal!

Sounds neat. It would be nice to see how it would scale to more complex scenes - I'm guessing it wouldn't very easily. But for matching speicific objects this could be very useful - especially characters. The occlusion test is obviously very neat.

Did he do something like have a set of functions to manipulate the standard P,I,N,s,t etc before feeding it to standard shaders? Can't wait for the website to be updated.

I don't think this is the first time he's done something like this - there was ment to be a previous 2003 Stupid RenderMan Trick where he saved out micropolygon grids, sent them to the GPU for processing, and really sped up renders. Sadly the notes aren't online as far as I can tell...

Is he at MPC at the moment? Hensons 3D completely shut down didn't it?

Simon

Bonedaddy
08-08-2005, 12:30 AM
I seem to recall all his variables were set with custom functions, i.e.

P=hb_get_point();

My memory's foggy on this point though.

beaker
08-08-2005, 02:56 AM
A few limitations though he mentioned. You can't assign different shaders to the object, only change the one that is already there. Also as Bonedaddy said, you have to raytrace everything.

kemijo
08-08-2005, 10:33 AM
There seems to be notes up now:

http://www.halbertram.com/interact/notes.htm

Just amazing.

-Sai-
08-08-2005, 03:56 PM
what next ? Online Rendering like Flame Artist?

imagine sitting with art director or whoever in charge of the shot and start tweaking the scene in almost real time.:thumbsup:

darktding
08-08-2005, 05:05 PM
I also think that the shaders are not your standard renderman type, they had to be altered a lot to accomodate for quicker feedback... like using only points of the surface... anyways forget about Hal.... check this out from pixar...
http://www.vidimce.org/publications/lpics/
this one is even crazier, and aparently Pixar has been using this tool for a while now... no wonder Jeremy Birns been telling me the tools he uses at work are far superior DOH!

beaker
08-09-2005, 01:51 AM
what next ? Online Rendering like Flame Artist?

imagine sitting with art director or whoever in charge of the shot and start tweaking the scene in almost real time.:thumbsup:This technology isn't really new. Just look at Worley Lab's G2(been around 2 years). It's just really cool to see this in Prman.

pearson
08-09-2005, 08:43 AM
Wow! 140 CPUs rendering one viewport! In just a couple of years all that power will sit in a conveniently sized box under my desk.


And I'll use it to surf the internet! :p

wurp
08-23-2005, 04:33 PM
Please no, I dont want some fool sitting next to me while doing my work, sometimes its nice that things takes some time otherwise we would never be left alone...



what next ? Online Rendering like Flame Artist?

imagine sitting with art director or whoever in charge of the shot and start tweaking the scene in almost real time.:thumbsup:

CGTalk Moderation
08-23-2005, 04:33 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.