level sets in houdini

[aracid]

Hey all

I was reading an article in animationmagazine.net about "the day after tomorrow" and they mentioned a technique called "level sets" to over come the effects of lumpy water.

how can somethin like that be implemented in houdini?

thanks in advance

aracid

[thekenny]

this is probably a good place to start:

http://math.berkeley.edu/~sethian/Explanat...et_explain.html

I think if you want to do this now you would have to do it yourself, or wait until DOPs come in the package. I for one am happy they have put more time into developing a solid robust solution, than just throwing it into the package with holes and bugs.

-k

[Jason]

thekenny's right, wait for DOPs. SESI have mentioned that they have implemented level sets in Houdini 8 (but by their other name, Signed Distance Fields or SDF's). We integrated our own level sets into Houdini during The Day After Tomorrow and they were invaluable especially in VEX (besides defining the water surface). We had a tool to convert our own format (an inventor variant) into i3d format which can represent SDF's just fine - and now you get the benefit of using SESI's optimized tools to access them.

The problems we have using i3d was that a lot of the time we would generate levelsets at different offsets than 0,0,0/0,0,0. Currently i3d's only support axis-aligned bounding boxes - i.e, or levelsets were often rotated and offset and we'd have to fit an axis-aligned i3d bbox to capture it all - which obviously required a re-sampling of the data - in effect degrading the data a generation.

Perhaps the i3d format or the sdf format or whatever SESI comes up with can support a full transform of the bbox. No doubt it can if its using SDFs in DOPs for RBD? (wha?)

Bring it on!

[Stremik]

I'm not shure if I understand the whole level sets idea correctly

but something similar can be simulated in Houdini with Surfsect SOP.

Isn't that so?

Vladimir

[thekenny]

try this one

http://www.exocortex.org/siggraph/2004-1.html

my previous one did not give you a sense of what you can do with them.

the Surfsect SOP is more of a boolen for patches. i don't think the results would be as useful.

-k

[Stremik]

I see. That does look preatty neat. Much cleaner than regular animated mesh. But! This kind of metaballish behaviour gives the model a feel like it was cut out of a solid piece of rock or whatever. What I mean is, when a character bends his arm the surfaces that are being pressed against each other rather being blended together like metaballs than squishing and deforming one another.

And another thing that popes in to my head is how would the texturing

be handled in these "level sets", when topology of the model changes all the time?

[Jason]

I see. That does look preatty neat. Much cleaner than regular animated mesh. But! This kind of metaballish behaviour gives the model a feel like it was cut out of a solid piece of rock or whatever. What I mean is, when a character bends his arm the surfaces that are being pressed against each other rather being blended together like metaballs than squishing and deforming one another.

And another thing that popes in to my head is how would the texturing

be handled in these "level sets", when topology of the model changes all the time?

14863[/snapback]

You're totally right - parametrizing the surface is nearly impossible right now. They're fantastically useful for technical purposes but this is one aspect that remains difficult.

[aracid]

sounds interesting

hey thekenny, thanks for the reply and links.

and thanks to jason, for the little explanation,

Its really exciting to hear that its being released on H8, so i'll definately gonna be waiting.

it kinda reminds me of goop,

goop

anyways,

thanks all

aracid

[Guest Ben Houston]

You're totally right - parametrizing the surface is nearly impossible right now. They're fantastically useful for technical purposes but this is one aspect that remains difficult.

14879[/snapback]

Actually it is quite easy to put a texture on these things. We did that for Scooby Doo 2. The method we used is described on this page:

http://www.exocortex.org/siggraph/2004-2.html

Basically we just represent the texture volumetrically too -- and then display it when the level set moves. It works out really well.

-ben

[thekenny]

Good stuff Ben. I saw FranticFilms is partnered with XSI. Do you guys use Houdini?

I had looked at levelSets a while ago to handle collisons for nonFluid based geometry. We ended up using a different approach all together. There are some interesting things you can do with them. Can you speak about FF's fluid tool?

-k

[Jason]

Actually it is quite easy to put a texture on these things.  We did that for Scooby Doo 2.  The method we used is described on this page:

Thanks for the link.

What I was really trying to talk about is trying to counteract stretching and compression of surface features for non-viscous water-like surfaces. Generating and removing details for watery surfaces can be pretty hard to do in a way that looks good, IMHO.

Your technique here looks great for thick, viscous surfaces like mud and tar. Very nice piece btw, well done!

[Guest Guest]

Good stuff Ben. I saw FranticFilms is partnered with XSI. Do you guys use Houdini?

I had looked at levelSets a while ago to handle collisons for nonFluid based geometry. We ended up using a different approach all together. There are some interesting things you can do with them. Can you speak about FF's fluid tool?

-k

14893[/snapback]

No we do not use Houdini. I do track incoming links to my website though... ;-)

Level sets for handling collisions with non-fluid based geometry. This is done in Eran Guendelman's 2003 SIGGRAPH paper:

http://graphics.stanford.edu/~erang/

I can't say too much about Flood except for the videos up on the Flood website and the pages that we link to at the bottom of the page. Currently we only have Flood working in 3ds Max -- although that may change:

http://software.franticfilms.com/index.aspx?page=flood

If you want to be added to the Flood beta list we have just drop me an email.

Thanks for the link.

What I was really trying to talk about is trying to counteract stretching and compression of surface features for non-viscous water-like surfaces.  Generating and removing details for watery surfaces can be pretty hard to do in a way that looks good, IMHO.

Your technique here looks great for thick, viscous surfaces like mud and tar. Very nice piece btw, well done!

14897[/snapback]

We just assume that our fluid is like normal fluid. Even if I were to create a block of colored fluid with crisp boundaries after mixing the fluid up a bit it will all sort of average out to the same color. This is our approach. We add in details via "morph targetting" -- a way of blending in features such as the facial detail of the Tar Monster. (Which is described in the 2004 sketch.)

[aracid]

hey all

could someone explain to me how the volume dop can be implemented ?

nothing major, but im just trying to get my head around the beta and thats one op im not getting right.

thanks in advance