A New Spin On The Cornell Box (animation)

[sasapp]

I built these animations to teach myself about a number of LW's awesome features. This first one combines image/texture-based lighting, radiosity, motion blur, and pseudo-hard body dynamics with Motion Designer.


http://lamar.colostate.edu/~sasapp/images/light-floor-box-01.gif
Click this link to see the full size MPEG:
http://lamar.colostate.edu/~sasapp/images/light-floor-box-01b.mpg

This one is just a reworking of the first one with better motion designer settings and a color cycling envelope on the color of the lighting texture.

http://lamar.colostate.edu/~sasapp/images/light-floor-box-02.gif
Click this link to see the full size MPEG
http://lamar.colostate.edu/~sasapp/images/light-floor-box-02b.mpg

I'd appreciate comments, but these are basically done, and I learned a lot in making them. I'm posting these just to share with this community that's been so helpful to me. Thanks for checking this out.

Shawnboy

[riki]

mmm looks good

[hrgiger]

Awesome. Are those just luminescent floor panels (or ceiling panels, can't really tell in this one since it's got that Lionel Ritchie dancin on the ceiling thing going on....) Rigid bodies look good too.

How long did that take to render?

[Hervé]

I am afraid Radiosity is not managable in a real production env. I mean going over 5 min. (for all the passes) rendering time is nothing good for animations.... I am doing a walkthrough anim of 2800 images.... how do you manage that with radiosity on ?,... baking all... well by the time it is finished I'll be dead.... anyway, Dimitri Saninoff showed me a super fast technique that looks like radiosity, but w/O the render times....


Check it here... (warning RAM intensive...!!)

http://www.savinoff.com/forum/viewtopic.php?t=19&sid=25a7144463789854118357c43f1ac81d

[riki]

I was having a meeting last week with a small start-up in Sydney that does maily toon based broadcast commericals (using maya). It's an off-shoot of a company that handles all the Warner Bros stuff in the Asian pacific region. I don't want to mention any names.

Anyway I was surprised when they told me that they never use GI or Radiosity in their work, just faking it. But the results looked perfectly acceptable.

[Hervé]

.... and you're right Riki, its like Savinoff told me, he has to produce..!!

BTW did you knew his work Riki ?

hervé

[hrgiger]

Originally posted by Hervé
I am afraid Radiosity is not managable in a real production env. I mean going over 5 min. (for all the passes) rendering time is nothing good for animations....

Of course, that really depends on who you are. Pixar says that some of their frames take three or four days to render. Mind you they have a renderfarm from hell.

[StuManFu]

A year ago I used to think GI use in production was a long way off. But we are using it increasingly. All thanks to the optomisations for the P4.

It all started off with basic diffuse shadow passes for CG into live action plates. And we're just completing a job that uses GI for pretty much everything. (Though I should add that it's pretty simple stuff).

And there was some pretty interesting stuff at Siggraph about real-time GI. But thats another thing...

Stuart Hall
Passion Pictures

P.S. 30mins is as long as we would like frames to take. But the average is about 5-15 mins.

[Titus]

Originally posted by Hervé
[B]I am afraid Radiosity is not managable in a real production env.... anyway, Dimitri Saninoff showed me a super fast technique that looks like radiosity, but w/O the render times....


I always been reluctant to global ilumination techniques for production. The light dome you see is a an old trick, you can do some "image base illumination" if you connect the light to an image and assign the closest pixel to every light light. I think Eki's overcaster is a derivation of the light dome.

Pixar is always looking for ways to optimize rendering times and hardly make use of raytracing, they do fake raytrace, even now with Nemo. So if they can produce high quality images without the need of GI, why us the mere mortals need it?

As an example of this. Pixar was rendering Nemo at the exhibition floor with a G5 machine and took like 4 minutes per frame.

[riki]

Thanks Hervé I'm just checking Savinoff's work now. Some interesting comments here.

[sasapp]

Thanks for all the input folks. I agree that GI is a bit beefy now, but if you think that this will be the case in 5 years, you're wrong. Look at the trends, GI is where its going and the hardware/software will be created to suit it. Any highend 3D package that doesn't have GI and sub-surface scattering capability at its core in a few years will be toast IMHO.

I do have a favorate radiosity faking technique though. I take whatever HDRI I want to use as a backdrop, use the LightGen plugin to create a large array of lights based on the HDRI, import these into LW, delete the lower half, and render with soft shadows. This really works well, but there's still something about the true radiosity renders that's very sexy.

Besides, I'm a chemist so I don't care if a little pet project takes too long. My main workstation is a Dual AMD Athlon 2GHz, plus I've got a render farm of three other old computers that when put together make quick work of even my toughest jobs.

I don't know if y'all have discovered this trick or not, but if you use AA and motion blur for your renders, you don't need a high sample setting in radiosity with reduced shading noise enabled (4x16 works just fine and is pretty fast).

Sorry to all you guys with unrealistic deadlines, but don't let the hardware on your desk fool you. It is coming...

BTW... the lighting panel is just a texture on the luminosity channel with an image I created in PS to set up the grid. This image is set to alpha mode with a "value" procedural layer below it. I just set the "value" to whatever brightness I want the light panel to be. These used a value of ca. 350%

[sasapp]

From what I know, Eki's overcaster stuff is all based on the spinning light trick, not dome lighting.

[sasapp]

One more thing. The first render used Interpolated mode and had 180 frames. The second used full on Monte Carlo calculation with 200 frames, but both only took a few hours to render on a P4 1.5 GHz. I don't see what the problem is. A half day is too much?

If you use just a few networked computers (even the cheap old ones) as a renderfarm this time goes down by several orders of magnitude. So using good techniques with Interpolated or Monte Carlo we're talking roughly a few hours to render even a broadcast quality animation of say a few thousand frames. If you're not getting these kinds of results, then you're doing something wrong.

[anieves]

maybe so, but try rendering a complex 800k+ poly count scene of a city scape and see what happens. ;)

[sasapp]

Very True. Point well taken.

[Titus]

Originally posted by sasapp
From what I know, Eki's overcaster stuff is all based on the spinning light trick, not dome lighting.

So a spinning light isn't much like a bunch of lights?

[sasapp]

The "spinning light trick" is based on a few lights set up to spin or sweep over a wide area of your scene during only a portion of a frame. So during a frame, your light spins around like crazy for 0.5 frames or whatever you set it up to be, then when you motion blur the frame, the renderer goes back and each aliasing render catches the light in a different position (ie. casting different shadows each time) and averages/smooths the result for the final rendered frame. So you get soft shadows and pseudo GI without all the fat. Eki's site has all this and more info so don't sit here and listen to me, go read about it for yourself.

Did you get all that? My point was somewhere around here... Oh, so the "spinning light trick" is not just a bunch of soft-shadow mapped lights. It relies inherently on the way LW's renderer does motion blurring, and thus also relies upon the way it anti-aliases the final render. Hope this helps.

[Titus]

Returning to the "bunch of lights" technique, I wrote a little script that create a dome with randomly separated lights. I was trying to implement a Hammersley deterministic method (trust me on this) but I'm stuck with a stupid for loop. This method (Hammersley) create a dome with perfect separated points no mather the number of lights you choose.

[Titus]

This is a faster render using spotlights with raytraced shados, I'm sure will be better and faster with shadow maps (I'm not sure of the memory consuption).

[Titus]

// by Felipe Esquivel
// version 0.1 9/8/03


@version 2.1
@warnings
@script generic

curScene;


generic
{

type = 1;
radius = 6.0;
lightNumber = 60;
totalIntensity = 1;
lightName = "dome";
intenScale = 1.0;
reqbegin("Light Doom");
reqsize(332,210);

c1 = ctlstring("Light Name", lightName);
ctlposition(c1,35,34,266,19);

c2 = ctlpopup("Light Type",type,@"Distant","Point","Spot","Area"@);
ctlposition(c2,67,63,232,19);

c3 = ctldistance("Dome Radius", radius);
ctlposition(c3,75,92,225,19);

c4 = ctlinteger("Number of lights", lightNumber);
ctlposition(c4,48,121,252,19);

c5 = ctlpercent("Total Intensity", totalIntensity);
ctlposition(c4,168,150,110,19);


return if !reqpost();


type = getvalue(c2);
radius = getvalue(c3);
lightNumber = getvalue(c4);
totalIntensity = getvalue(c5);


reqend();

var lname, parNull, parId, i, lin=1;
var red, green, blue, mx, intens, x, y, z;
curScene = Scene();
(parNull) = curScene.getSelect();


for(i=1; i<=lightNumber; i+= 1)
{ if(type==1)
AddDistantLight(lightName + "_" + i);
else if(type==2)
AddPointLight(lightName + "_" + i);
else if(type==3)
AddSpotlight(lightName + "_" + i);
else
AddAreaLight(lightName + "_" + i);

parId = parNull.id;
TargetItem(parId);
HController(2);
PController(2);

theta = random(0, 6.2832);
eta = random(0, 3.1416);
x = radius * sin(theta)* cos(eta);
y = radius * sin(theta)* sin(eta);
z = radius * cos(theta);
if (y < 0)
y = -y;



Position(x, y, z);
LightColor(1, 1, 1);
LightIntensity(totalIntensity/lightNumber);
}
}

[Titus]

Sorry, Do I'm bugging too much in the Finished Gallery?