PDA

View Full Version : Strauss Lighting



LaBasX2
01-13-2006, 09:45 AM
Hello,

in the 3dmark06 interview on beyond3d they are saying that the new version of the benchmark will use Strauss lighting.
This probably refers to a paper called "A Realistic Lighting Model for Computer Animators" by Paul Strauss. Unfortunately the paper is from 1990 and costs 20 Dollars and that's a bit much for just being curious. Does anybody know somehting about that lighting technique?

Interview (http://www.beyond3d.com/interviews/fm05/p2/index.php?p=01)

Thanks!

dorbie
01-13-2006, 10:13 AM
A Realistic Lighting Model for Computer Animators

Source IEEE Computer Graphics and Applications archive
Volume 10 , Issue 6 (November 1990) table of contents

Pages: 56 - 64
Year of Publication: 1990
ISSN:0272-1716
Author Paul S. Strauss

Publisher IEEE Computer Society Press Los Alamitos, CA, USA

ABSTRACT
Because of the problems inherent in modeling the physics of light transmission, all lighting models used in computer graphics are approximations. Each model provides a different balance between image realism, model complexity, and runtime efficiency. A model that balances these factors and is easier than its predecessors for inexperienced animators to use is presented.

dorbie
01-13-2006, 10:48 AM
I'm looking at the paper, it starts with a quick survey of popular lighting models (Phong, Blinn, Torrance Sparrow, Cook Torrance & modifications like additional colors) and their problems w.r.t. physical accuracy. The paper then presents a new formulation which uses 5 parameters to describe a surface and then presents the equations for their use in calculating lighting results.

The main difference in the diffuse calculation seems to be the inclusion of a surface roughness/smoothness factor (also another parameter called 'metalness') to empirically derive diffuse reflection values. The underlying calculation still seems to be N.L. It's modulated by the diffuse reflection term(s) AND surface color.

The specular equation is still basically V.H but the exponent is derived from the same parameter that one of the diffuse modulations is, i.e. surface smoothness and the result modulated by reflectivity where reflectivity is also derived from other parameters (including transmission and that earlier diffuse term based on smoothness) clamped to one. There seems to be a fudge curve or two in here to produce reflection angle peaks consistent with Fresnel's equations.

The overall goal of the work seems to be to ensure that the totals for reflected specular and diffuse light (and transmitted light) are physically consistent and match other physical phenomena (Fresnel's), while at the same time provide just a few intuitive parameters to describe the surface. i.e. you can't max out specular and diffuse because that's not what happens with real surfaces, or to put it another way, a surface will not reflect more light than is incident upon it.
The paper ends with some notes on physically based transmission and refraction in global illumination with physically based refraction angles.

It probably looks better, but maybe not a whole lot better when you consider content like textures etc. I suspect it's pretty good for glass & translucent surfaces because of the inclusion of Fresnel's equations for specular & transmisison factors. It's probably also good for global illumination because of the attempt to produce physically accurate totals something Blinn's is hopeless at. It may also counter the 'plastic' look of vanilla Blinn surfaces making it easier for artists, for example, to get a realistic metalic look quickly.

LaBasX2
01-13-2006, 12:45 PM
Dorbie, thank you very much for this detailed summary!

The lighting model looks pretty interesting. The first 3dmark06 screenshots are quite impressive, I wonder if they are using Strauss as default lighting model.

I'll try to get the paper from my university to have a deeper look.

Thanks again!

zed
01-13-2006, 04:08 PM
Originally posted by LaBasX2:
The lighting model looks pretty interesting. The first 3dmark06 screenshots are quite impressive, I wonder if they are using Strauss as default lighting model.i only see the one screenshot (of town behind a body of water) which doesnt look to good (very flat)

on the subject of lighting ive been thinking for years that seperating diffuse + specular aint good, are there any lighting models that do away with them?

Korval
01-13-2006, 05:14 PM
on the subject of lighting ive been thinking for years that seperating diffuse + specular aint good, are there any lighting models that do away with them?Not good? That's how light works.

It all goes back to wave/particle duality. When light interacts with a surface in a wave-like manor, it is absorbed and re-emitted. Because the emission has no bearing on the direction of the viewer (the photon is emitted in an even propagation wave) the view direction is irrelevant. The light source direction is important, as this determines the general quantity of light that hits the surface. Hence: diffuse lighting.

When light interacts with a surface in a particle-like fashion, it bounces off (like a particle). The strength of the reflected light depends entirely on the angle of the viewer (and of the light source, of course).

You get different colors because surfaces will only exhibit diffuse lighting for certain wavelengths of light, and all others will be specularly reflected (or, for really strong frequencies, will cause a chemical reaction). The emitted light is often of a sequence of wavelengths based on the molecular structure of the material in question. Reflected light is just what wavelengths are left over or did not interact properly with the surface.

dmy
01-14-2006, 06:41 AM
specularity and diffuseness are abstraction/models, created by us, in the strive to get something simple to work with.

light do not work as in our model, but it's the model wich only loosely resembles how light appears, not works...

for a photon, a surface has not properties such diff or spec, it simply interacts with what it will hit on its path: then, the composition of all the results born from those interactions, make the appearance of spec/diff/whatever.

zed
01-14-2006, 11:27 AM
ive studied first level vulcan so was able to decifer your post korval (just kidding, :D ). Personally im more concered about something that looks good than something that is physically accurate ( related to i posted here a couple of months ago 5 screenshots of different lighting models to gauge ppls opions (*) )

(culled piece about HDR lighting here)

(*) actually if anyone has a link to a paper/page that shows the same scene lit with various light models side by side i would love to see it, nowadays with shaders we have the flexibilty to implement different methods.

SirKnight
01-14-2006, 12:35 PM
Originally posted by zed:
ive studied first level vulcan so was able to decifer your post korval ... Noshau u'dvinsular tum-vel ohrom heh yeht-urgam'es - nu ri aitlu ne'au dvin-tor. :D

-SirKnight

Korval
01-14-2006, 05:31 PM
specularity and diffuseness are abstraction/models, created by us, in the strive to get something simple to work with.Yeah, the whole wave-particle duality, surface interaction stuff, all that stuff I was talking about in my post. I just made that up. :rolleyes:

Go learn some modestly high-level physics/chemistry. When you understand the mechanisms by which light interacts with the electron cloud of atoms, we'll talk. Until then, no.


Personally im more concered about something that looks good than something that is physically accurate Physically accurate looks good. So it makes sense to do what physics does. Or, at least, try to approximate it.

The most persistent problem with most surfaces isn't the lighting model. Even Phong or Blinn can look quite good if the surface details are there. No, the problem is in putting in all the textural detail that most surfaces have.

dmy
01-15-2006, 09:07 AM
@korval:
you are misunderstanding me, i've never supposed you made up the things you said 'cause i know my physics too,
simply you didn't understood what i was talking about. if i offended you, i apologize: it wasn't intentional.

you replied to zed:

Not good? That's how light works. while referring to:

... seperating diffuse + specular aint good ... what i meant was that physics models light interaction with matter in a more complex way than
diffuse + specular + whatever (as gl does) because it encompasses many more macroscopic/microscopic behaviours
of such interaction, and it does with great precision, since it don't use our ('our' in the sense of 'us graphics programmers')
simple tristimulus RGB paradigm.

zed
01-15-2006, 12:08 PM
Noshau u'dvinsular tum-vel ohrom heh yeht-urgam'es - nu ri aitlu ne'au dvin-tor.somehow as someone who purports to be from the round table of camelot i dont find this surprising. :)


Physically accurate looks good. So it makes sense to do what physics does. Or, at least, try to approximate it.in real life u tend have lots of light bounces thus things aint as defined as the user would like. hmm cant explain.
ok
say u have a person walking down a corridoor in an actual photo s/he will cast no shadow (perhaps just a general slight darkening of the surrounding area)
now in a game im sure 90% of users would prefer a distintive humanlike shadow (ideally sharp under the characters feet fade to blured the further up the body)
another example (since i live in a place that overlooks a body of very still water ive been observing it (+ taking pictures) over the last few months)
- reflections in water look good, check out farcry etc, im sure 90% of ppl will agree they add to the scene. but in reality, its rare to have such distinct reflections (with ripples). in reality u need ideal conditions eg wind less than 5km/h. otherwise u will not see the tree reflected in the water but an indistinct colorblur.

perhaps i can use the term u mentioned a while ago (something trough or something) where the closer to realism the effect becomes the worse it looks.


The most persistent problem with most surfaces isn't the lighting model. Even Phong or Blinn can look quite good if the surface details are there. No, the problem is in putting in all the textural detail that most surfaces
have.whilst i do somewhat agree with you, look at some radiosity screenshots (without any textures at all) hmm didnt i just contradict the above point :)

Korval
01-15-2006, 03:50 PM
what i meant was that physics models light interaction with matter in a more complex way than
diffuse + specular + whatever (as gl does) because it encompasses many more macroscopic/microscopic behaviours
of such interaction, and it does with great precision, since it don't use our ('our' in the sense of 'us graphics programmers')
simple tristimulus RGB paradigm.No, the real world is doing exactly that: specular + diffuse. The difference is that the real world is doing it across moles (http://en.wikipedia.org/wiki/Mole_unit) of photons and moles of surface atoms and molecules. The precise description of both these molecules and the geometry of them relative to one another also helps determine the outcome. We lack the ability to describe either of these in a convincing way, but we can approximate them pretty well.

The problem isn't having diffuse and specular terms. The problem is not being able to accurately describe the surface to significant detail and the complicated equations that make up photon-surface interactions. Therefore, the problem is in what equations you use to determine the diffuse and specular interactions, not in adding them together or in having the distinction at all.


whilst i do somewhat agree with you, look at some radiosity screenshots (without any textures at all) hmm didnt i just contradict the above pointWhat does radiosity have to do with "specular + diffuse"? Radiosity is a mechanism for determining the quantity of light that interacts with a surface, given light reflection/emmision from other surfaces. "Specular and diffuse" are the mechanism by which light actually interacts with a surface and the mechanism by which one determines how a surface reflects/emits that light back into the scene.

These are two entirely seperate problems. How well one does global illumination has nothing to do with how well one does light/surface interactions.

Obli
01-16-2006, 09:12 AM
Originally posted by dorbie:
The main difference in the diffuse calculation seems to be the inclusion of a surface roughness/smoothness factor (also another parameter called 'metalness') to empirically derive diffuse reflection values. The underlying calculation still seems to be N.L. It's modulated by the diffuse reflection term(s) AND surface color.This recalls me Oren-Nayar... how the two compare against each other?

dorbie
01-16-2006, 09:40 AM
How well one does global illumination has nothing to do with how well one does light/surface interactions.Korval, I'm sure you'd like to rephrase that, the light/surface interaction underpins any global illumination system, and is apparently one of the motivations for the Strauss lighting model.

Global illumination results are meaningless if your surfaces are not approximately physically correct or worse not even physically plausible.

Korval
01-16-2006, 10:48 AM
Korval, I'm sure you'd like to rephrase thatOK, in terms of the overall result, they both matter. But they are two different things; how one computes one does not influence how one computes the other. It may influence how one wants to compute the other, but the two pieces of actual code do not directly touch one another.

dmy
01-18-2006, 05:06 AM
ok, i got your point: statistically, light-matter interactions can be modeled exactly as diffuse+specular... good to know, thanks k.

V-man
01-18-2006, 04:54 PM
Originally posted by Korval:
[QUOTE]No, the real world is doing exactly that: specular + diffuse. The difference is that the real world is doing it across moles of photons and moles of surface atoms and molecules. The precise description of both these molecules and the geometry of them relative to one another also helps determine the outcome. We lack the ability to describe either of these in a convincing way, but we can approximate them pretty well.The real world does transmission, refraction, diffraction, ...


Originally posted by Korval:
[QUOTE]It all goes back to wave/particle duality. When light interacts with a surface in a wave-like manor, it is absorbed and re-emitted. Because the emission has no bearing on the direction of the viewer (the photon is emitted in an even propagation wave) the view direction is irrelevant. The light source direction is important, as this determines the general quantity of light that hits the surface. Hence: diffuse lighting.I'm a little rusty on physics but I think diffuse doesn't signify wave behavior and particles doesn't signify specular behavior.
Waves can bounce giving specular behavior. Perfect reflection with no energy lose.
Particles can hit an atom, lose some energy and bounce back.

The wave theory comes from observing diffraction (interference patterns) and other cases.
The particle theory comes from observing the photoelectric effect and other cases.

So what are people's opinions on this Strauss lighting equation. Must have or nothing special?

Korval
01-18-2006, 05:16 PM
I'm a little rusty on physics but I think diffuse doesn't signify wave behavior and particles doesn't signify specular behavior.OK, technically you're right, but I didn't want to give an entire chemistry/physics lecture in the middle of an OpenGL forum. Despite the particular inaccuracy you're referring to, it does correctly explain how diffuse and specular emission/reflectance work at the atomic level.


So what are people's opinions on this Strauss lighting equation. Must have or nothing special?Must have? No, probably not. But it seems to be a pretty reasonable fairly high-end surface interaction scheme. It also seems pretty flexible, perhaps with the ability to do metals pretty well.