Slashdot Mirror
SIGGRAPH 2001
Morgan McGuire writes "SIGGRAPH 2001, the graphics industry's main scientific conference and gathering for artists, film producers, researchers, and game developers, just ended.
I wrote up my experiences as a game developer/researcher at the conference for flipcode." Lots of stuff for those of us who wish we could go every year and see the pretty pictures. Hits on Shrek, Monsters, Inc. and a variety of new techniques floating around.
In addition to the wealth of knowledge you can get from the conference, the contacts you can make in the industry are worth the price of admission. Where else can you get a class taught by Jim Blinn?
Where are we going and why am I in this handbasket?
In re: the "panel" (which was not) between gamers and scientists... I work with visualization (*yawn*) and other "scientific" apps... though I was the first to bring 7 megapixel (150" diagonal) quake3 to our lab ;). The truth is, the annoying thing from a research perspective is that there _have_ been huge strides made in the last 5 years or so, thanks to the gaming market (which in terms of people is probably 4 or 5 orders of magnitiude larger, while monetarily "only" 3-4). The agravating part is that despite these leaps, they're not completely focused on the things that "matter" in research. E.g... your average quake3 map requires rendering less than 100 polygons (for the background, at least... throw in another 1000 max for characters) with huge textures... aka... fill rate. Your average scientific display requires 100,000 polygons (minimum... most of the data I see is between 300,000 - 5,000,000 (that's million, not thousand), but with no texturing at all. The difference in a consumer card (Geforce 2, say) and a so-called "professional" card (Quadro 2) is now only a few hundred (ok, maybe 500) dollars... nowhere near the few thousand it used to be. But it's still there, and it doesn't look like it's getting any closer. Still, that's a huge improvement over even a year or two ago. My boss has a FireGL card in one of his machines... it can do 10 - 20 times the number of polygons per second of a Geforce 2 GTS... but in UT, at 1024x768... it only gets about 30 fps. But that card was 1500 dollars just a year ago... and three years ago, it would've required an SGI for 10K+ to beat a consumer card like that in polygons.
-- Is "Sig" copyrighted by www.sig.com?
What I find really sad is the fact that now the internet is mainstream, web and video technologies are available to anyone that wants to play with them, and now we're talking about THE graphics show and multimedia experience, and there's still no webcasting of events or a centralized press release/video depository of events.
:) ) might need just that little boost.
/web technician and voala... if Newtek pulled it off for their booth (and I am sure they aren't the only company who did that), why can't siggraph itself do something like that? Just an Idea.
You have to go to EACH companies's siggraph website (that's when they take or have the time/staff to do so and not everybody running around fixing last minute issues).
I was locked on Newtek's streaming event for a while and was thinking to myself "god, that would have been so cool having a reporter on the floor going from companies to companies, webcasting all day, and a place with different archived video of reports for me to check, by companies" you know, something SIMPLE compared to the whole organisation needed to make such a show a reality.
Don't get me wrong, this isn't a siggraph bashing, I'd LOVE to attend the show once in my life, but also, since it's so BIG and all the major announcements for the next 6 months are happening there, and the fact that industries that pushes visual technologies are represented there in an "international show", I don't understand why in 2001 we still don't have that simple technology available to make this event even bigger.
To get the most attendee possible without people thinking "I'll stay home and watch the show"? I don't think it would cut in the attendance, because people that WANT and can afford to go there won't be satisfied with a webcast, but people who WISHED they were there at least will have a glimpse...Probably even potential people that might want to go the year after because they can feel the atmosphere (with good reporting
Cost? That also I wouldn't agree, one word: Advertising... want traffic? tell me that wouldn't bring any traffic to their site? They could get their extra troubles easily refunded with that, plus even generating more money. The siggraph.org web site looks so.... dunno... dull maybe? Pictues/companies PRs/video would give it more life.
Anyways I'm sure they could have pulled this off easily, one webcasting server, 1 camera man, one reporter and one video editing/archiving
--- Metamoderating abusive downgraders since my 300th post.
Mechanical trick? Alien virus? Nope, the black oil is a ferrofluid, a suspension of regular oil and magnetic micropowder.
A quick glance at the site was rather informative. Hey! Do an article on ferrofluids or something. They look like they'd be incredibly fun to play with.
Where the wind blows, the tumbleweed goes.
Was just me, or was Siggraph much smaller this year? I've been going for over 10 years, and just seemed as though there weren't as many people and vendors. Side Effects (Houdini) wasn't there, and Disney also pulled out at the last minute. I think everyone I ran into mentioned it was a much calmer show than previous years.
Also - the level of technical advancement seems to be leveling off. In the past, my jaw always dropped at least once during the show. The NVidia/Square realtime Final Fantasy was cool, and there was a nice IBM 600dpi monitor, but that was about it. Outside of a few things (ExLuna, for one) no huge software releases, either. Mostly incremental improvements.
Perhaps it's to the point where the technology is getting "good enough" (relative term, I know...) The only things that really made my jaw drop was the content itself. In the Electronic Theater, I'd see a really good film, then the credits would list just one person... Jaw drop. Pretty amazing how far it's all come.
I can't believe that the Slashdot geeks are tarnishing their name so much.. no one has mentioned the "2001 in 2001: How a Film Inspired Our Future" session on Friday.
We had Robert Abel, the inventer of slit scanning (remember the 4th Doctor Who opening?) who produced the graphics for the final 1/3 of the movie (the Jupiter sequence). There was Syd Mead, the person who designed the look of Blade Runner and 2010. There was also Peter Hyams, director and photographer of 2010 and other (crap) films (End of Days), and Dennis Muren of ILM that helped create some of the real time compositing stuff for A.I.
They talked about Kubrick a lot, but had some problems staying on topic. Syd Mead gave a great little intro to his vast array of work (including the design of cars and plane), and Bob Abel actually gave some major explainations of how different parts of the movie were done. It was my favorite thing there this year.
I'm not a graphics guy (I am a Linux admin that works for a department that teaches Maya and 3DM, so I get a free ride, it's in the budget), so SIGGRAPH isn't my major conference (that would be OSCON), but I did enjoy it.
Oh, on a final note, there were several sessions about public policy. There was a lot of talk about Dimitri, DeCSS and other IP issues that everyone here would have been proud of. ACM and SIGGRAPH are solidly against the DMCA.
"Yes.. no matter what the culture, folk dancing is stupid." -MST3K
This is also of interest to computer game players, as the techniques they have developed here will apply to NPC's nicely now that computer firepower is catching up to our ideas. WETA has been working on this software for 2+ years now, ever since they began work on LOTR.
Basically, each character on the battlefield has a node-based brain full of IF/THEN/OR type junctions. They began by showing a single human character wandering around a maze of walls and blocks. In one corner was a view of what the character was 'seeing' through it's eyes, head-bob, swaying and all. When the character got near a wall, it would stop, look around, turn and walk in another direction. All this was being done by seamlessly morphing between different sets of motion capture data. This in itself was a very nice thing, as it's usually very hard to do that completely automatically and have it look perfect.
The next step was collision detection. They put multiple characters in the maze, and they would all avoid each other, smoothly, not abruptly. After that, they put a guy on hilly terrain, and he seamlessly morphed between uphill walking and downhill walking as the terrain called for it. (heavy slow stepping uphill, shifting weight backwards downhill...)
This particular character had about 500 nodes in his 'brain'. The characters used in the movie have around 5,000 nodes. Why? Here's what else they do:
The characters can be programmed to charge at each other, and when two characters from opposing armies see each other close by, they run at each other and break out into motion capture sword fights! The sword fights are choreographed so that the mocap from one character matches the mocap from the other, so when one guy swings, the other blocks and it all works! And all of it blends together pretty damn smoothly, from one swing to another. Furthermore, each army has different mocap data. One army tends to fight with Eastern combat techniques, and another uses more European style swordplay, and they make all this work together, automatically.
On top of that, they have a randomizer. The characters will vary in size and attributes, with low and high limits being set, or in the case of shields and accesories, an on-off randomness that will determine if each instance wil have that configuration. Short characters walk faster to keep up with the taller ones, and there are several different walking datasets for more randomness. They showed an 'adventure party' of 6 Orc type characters, first without the randomizer and after with.
When I was working at Digital Domain, someone came up with a Who generator for The Grinch that made random Whoville citizens with different hats and shirts and sizes and whatnot. This is like the Who generator times 100. Really elegant stuff for in-house software. Also, it's very fast, and I believe at the show it was running under Linux. They had many characters each with a complex brain all running around at once, in realtime. Lots of characters morphing between mocap data, walking up and down hilly terrain and attacking each other, all automatically. And of course each character can be customized or scripted to do specific things to get more control in the foreground.
This demo was given at the SGI booth, the big one just as you come in. They had lots of demos from different companies throughout the day.
The second coolest thing was some new tracking, panorama and cleanup software from a company called (I think) 2d3 or some such thing. I'll mention it to the general crowd here because it has usefulness outside of my rarified industry. They had one piece of software that would take a recording of a video camera being panned around in a circle on a tripod, and turn that into a 360 panorama. Furthermore, you could tilt up and down and it would just get whatever you shot, by continuously tracking your motion to see where you're moving. on top of that, it would use many frames of video to assemble each section, so it would have an interpolative effect and you'd get a lot more resolution out of each section of the pano than you normally would have gotten in the camera, because it's assembling multiple samples of the same thing, grabbing detail that had fallen between the pixels in one frame from other frame while you were panning across the same area. Stuff that we've all dreamed about, but never thought anyone could actually pull off. Just set your shutter speed way up to avoid motion blur and have at.
Other than that, there were way too many crappy mocap booths and 3D printers. At least the 3D printing/rapid prototyping thing is becoming cheap(er) and more common. Soon I'll be able to print out props for costumes at reasonable cost, and then airbrush paint them.
On a semi-related note, if you were at the BLUR studios party, you can check out my pictures of the firedancers here: http://www.mikemassee.com/firedance/
--Mike