Slashdot Mirror
Perfect Digital Skin
ILMfan writes "BBC Technology is describing a new graphics algorithm for creating perfect virtual skin. This technique by graphics wizard Henrik Jensen (the guy who invented photon mapping) is already being used in movies (it was used on Gollum in Lord of the Rings, and it will be used in the soon-to-be-released van Helsing movie). And perhaps more exciting is that several game companies are planning on using it for their next generation games. So John Carmack are you listening? Any chance this can be included in DOOM3? Of course there are endless other opportunities for virtual humans with perfect skin :-)"
Henrik Wann Jensen also got a technical Oscar earlier this year. This work is actually quite old. See the original Siggraph 2001 paper here.
Don't Panic
http://www.tomshardware.com/graphic/20040504/ati-x 800-04.html
It's gonna make everything look that much sweeter...
-Christopher Wu
http://www.christopherwu.net/
this /is/ subsurface scattering that they're talking about.
Oops... links from ATI site MUCH faster:
s s-Demo-v1.0.mov
e -Demo-v1.0.mov
http://www2.ati.com/misc/demos/ATI-X800-DoubleCro
http://www2.ati.com/misc/demos/ATI-X800-Subsurfac
-Christopher Wu
http://www.christopherwu.net/
The technology is already available for games, check out the subsurface scattering demo from ATI: http://www.ati.com/developer/demos/rx800.html
Actually, depending on how long it takes to crank out a flick, it might be cheaper to pay the engineers. Remember, girls are getting paid anywhere from $200-$3000 PER SCENE, maybe 2-3 scenes per day. Add in pay for the guys, cameraman, location rental, etc. (all of which could be pennies based on the current "gonzo" porno trend), and you are talking upwards of 5-10 grand a day. 4 engineers @ $30/hour * 8 hours = $960 / day. Even if it took them a week you might still be saving money.
Green-voting, republican-registered, socialist-libertarian.
The most ridiculous part to me of lens-flare is that originally, it was to be avoided at all costs since it interfered with the suspension of disbelief (ie. it reminds the viewer that they're viewing something seen by a camera, not them), but somehow it got absorbed into the grammar of cinema as being cool. Videogames, not actually using a lens in the rendering process, were immune to the effect, but labored hard in efforts to reproduce it.
Not A Sig
This is also mentioned in his book, (2001), which I highly recommend to anyone interested in raytracing. It's short and about as easy to understand as photon mapping could possibly be.
He has a lot of stuff on his webpage, too, including videos of computer-generated smoke, light through translucent materials, and a good global illumination demo.
For a simpler explaination of what this is all about, there's a photon mapping entry at wikipedia.
-jim
Actually, this technique uses sub-surface scattering to accurately model the light transport characteristics of human skin.
Henrik has been a pioneer in developing efficient techniques for representing BSSRDF (bidirectional sub-surface scattering distribution functions).
This paper that he published in collaboration with other notable people at Stanford was among the first to describe methods of calculating the effects of sub-surface scattering.
No he didn't make Gollum. Jensen was the main researcher of this new subsurface scattering technique when he was at Stanford (he is now in San Diego). As was mentioned earlier this was published in a SIGGRAPH 2001 paper, so at least the research might date as far back as early 2001, late 2000. Jensen along with Steve Marschner and Pat Hanrahan got a SciTech Academy Award earlier this year for it (though Marc Levoy was omitted).
The original implementation used raytrcing to achieve the effect, to slow for actual production work. Some people from ILM spotted the paper and decided to implement in a way more friendly to production. Originally it was going to be used for Ep. 2, but the research wasn't completed on time. The first time it was applied was for Dobby in Harry Potter and the Chamber of Secrets. Christophe Hery even presented a Stupid RenderMan Trick using shadow buffers that could be applied to SSS. Around that time Ken McGaugh and Joe Letteri left ILM (tough they were involved with this research at ILM) and joined Weta Digital to work on the Two Towers. Consecuently, Hery, McGaugh and Letteri also received a SciTech Academy award this year for finding a way to implement Jensen's SSS in a production environment.
Subsurface scattering is quite old - I learned about it in my graphics classes, and I've been out of school since 1996... here's a 1993 paper on it.
He points out on his web page "Photon mapping is quite good at simulating subsurface scattering, but it becomes costly for highly scattering materials such as milk and skin. For these materials it is better to use a diffusion approximation. The diffusion approximation is much faster than tracing individual photons, and it is simple enough that a BSSRDF can be formulated."
Here's a BSSRDF from a google search.
Looking at that image, there are a couple of cues that scream artificial to me...
1) The reflection off the lips is too bright
2) There's something wrong with the shadow of the nose
3) The tip of the nose is too reflective.. in fact all of the areas that are in direct light are too reflective
4) The proportions of the face are off slightly (or they are too perfect and symmetrical to be real)
5) It may be the render resolution, but the imperfections of the cheek are too lumpy
But it is a heck of a lot better than a couple of years ago.
"44 quintuple Q"? You must be a geek. You don't know how bust size is calculated. The number is the circumfrence of the chest excluding the breasts. The cup size is based on the difference between that measurement and the circmfrence of the chest including the breasts.
So, your pr0n goddess would be fat as well as large breasted. In propper digital pr0n she'd be:
Bust: 36ZZZ
Waist: 17
Hips: 35
With no gag reflex and an abdominal cavity with the storage capacity of an oil tanker.
Boobies never hurt anyone. - Sherry Glaser.
BTW, literally "perfect" skin would mostly resemble Campbell's Cream of Bean soup.
Seeing bad movies only encourages them. Watch responsibly
I saw that same photo in a national geogrpahic magazine about how far from perfect skin we are from almost a year ago (and they seemed to indicate that it was obvious that it wasn't human skin which is in contrast to what the bbc implies).
The Television Wiki
Perfect skin is nice at all but it is very unrealistic skin. That is why computer generated characters still fail the reality test to the eye .. or at least one reason. Any natural thing is somewhat imperfect and not quite asymmetrical for instance take the left half of a face and mirror it in a graphics program so it is perfectly symmetrical. it looks ...odd.. You may not be able to quite put your finger on it ( if it's well done ) but it doesn't look real.
âoeTolerance applies only to persons, but never to truth. Intolerance applies only to truth, but never to persons.
Of course not. Photon mapping has not been done in realtime on standard machines (although GPU-based algorithm has been done, it's not realtime).
"Politicians and diapers must be changed often, and for the same reason."
BTW, here's the technique. It's really simple actually:
Before rendering each frame:
- sample the object with points on its surface;
- solve the illumination of each sample point (to get the amount of light falling on each)
- save this data somewhere.
During rendering:
- For every point being rendered, look for nearby samples (several techniques could be used for this, the paper used an octree structure, but my first implementation used just a flat 3d bucket sort of the samples)
- Average the values of all nearby samples, with some kind of falloff (the paper does the right physical think, using an exponential function--my first implementation used a cubic function that approximated the exponential while falling off to zero at a finite distance)
The value you get approximates the amount of scattered light. That's all there is to our 2002 paper, really. The rest is math, justification and embellishments. There is a lot more to rendering skin than subsurface scattering though, like layering and good texture maps, plus usability if you have tens of artists using the tools.
And yes, the technique could be patented, but it would very much suck if we had done that. My implementation is closed, since I wrote it at PDI on our propietary renderer, but as you see, it's simple to implement anyway.
Juan Buhler
(coauthored Henrik's 2002 paper)