D. Kirk and Prof. Slusallek discuss real-time raytracing

I completely disagree. QED is not really a model suitable to realtime and/or offline rendering as general tool.
C'mon..we don't want to simulate that kind of interactions, most of the time (99.99999..%) it would be a complete overkill and moreover QED is a complexity monster!

Well..I'm waiting for your implementation but I'm very doubtful..

 

However, the whole idea of raytracing being the full thing is still wrong, as it's just another hack...

 

It's always going to be a "hack" as long as it's not real. As long as it's something that we ultimately have to look at through a 2D display of variable resolution.

 

What about Radiosity?

Hmm, makes me wonder if radiosity alone could be calculated for a scene and blended as an additional lighting pass for rasterising hardware.

 

Heard of the uncertainy princple? also being a chem undergrade and actually do quatum mechnical modeling in another screen we still have a long way to go

 

Just use Monte Carlo techniques Of course that won't make it faster.

But back on topic. RT doesn't solve radiosity and it doesn't solve diffraction and refraction (with regard to the frequency contents of light).

Cheers
Gubbi

 

well i disagree with your disagreement

i´m well aware of all the problems qed brings with it, but if you really want to calculate lighting you will need qed. and you don´t have to calculate every possible subatomar interaction that possibly can happen. of course, if you do this you´re right in approximation and interpolation again, but well, whats the point of calculating in a microscopic environment when you only can display macroscopic things. i know that it´s a complexity monster, and i think i said that no one who really knows what he is trying to do should mess with that stuff. maybe that includes me too, but who cares i´m working on this lighting model a half year now so i´m not going to stop.
about the calculation overkill: i know what you mean, of course you have tons of possible ways light could move from point a to b, but what you actually do if you calculate with qed is finding the paths that are relevant and ignore the rest. or does it make sense to calculate a path where the probabilty is 0.001% that a photon takes it.

i´m not going to fully disclose my algorithm now, but it builds upon g-buffer rendering and impulse response calcualtion via convolution (each point on a surface is represented by a FIR digital system) and some photon-phase-prediction stuff. i can´t describe more now cause my technical english is way below that level and i don´t think you guys will be happy if i write in german, but i will piece together a paper wherer i fully describe the algorithm.

well enough for today, this is indeed a very complex subject and i don´t think it can be put into one single post on a forum...

 

Well..I'm a physicist...freel free to pm me

 

i´ll do that.

have you ever tried implementing something basing on qed or do you just think the attempt is ridiculous?

 

Never. QED was not my field..and I'm not doing physics research anymore (I work as game developer now..). But I hope I can have a grasp of your method

ciao,
Marco

 

Posts away here in public I'm sure if people can't grasp everything they will go look up what they don't know.

 

Really? Don't seem to recall it happening all that often when it comes to the graphics field

 

Thats cause theirs no good place to download all teh papers for free :/ theirs heaps of stuff on quatum mechanics around its surprising.


Now peopel are saying its hard for the artists to get the effects they want with Ray tracing how about trying to get the artist to recreate the world down to the single attom for diehaerte's idea

 

1) Artists are physicists.
2) It would be impossibly complicated.
3) It would be impossibly complicated.
4) It would be impossibly complicated.

 

random point that i'm not sure has been mentioned yet:

ILM don't raytrace for movie special effects, apparently they just use scanline rendering and limit themselves to just one light + ambient occlusion. I know ILM aren't the gospel for CG but they are pretty experienced, if they don't use raytracing for offline rendering then surely it should say something about whether it's useful for realtime.

 

Are you sure about that? Where did u read it?

 

A talk from a programmer at ILM, i was pretty skeptical myself but that's what he was saying. They just don't have the time with their pipeline to do really complex lighting calculations, they use just as many hacks as you'd expect games to.

It's possible that for *really* complex scenes they might switch to different lighting models but for bread and butter shots they don't.

 

This was in their 2002 Siggraph presentation:
http://www.renderman.org/RMR/Books/sig02.course16.pdf.gz

They are indeed using Pixar Renderman, and do the lighting with enviroment maps + ambient occlusion; however, this is for VFX shots where they have to composite CG elements into a live background plate.

The more interesting is reflection occlusion (also in the docs) - they have to raytrace this as a separate pass, and it's view-dependent, but they still prefer it instead of full raytraced reflections...

 

you don´t have to recreate the world in single atoms, you can extrapolate the results you would get if you really applied quantum mechanics to every single atom to a larger area (i.e. one pixel). you just need to know how this material behaves for some arbitrary 'signal', something that is called impulse response.

 

Ah, QED, of course!

We won't get realistic lighting until we get accurate modelling of the particles that the photons themselves pop in and out of the vacuum! Obviously this will be necessary for accurate modelling.