12 years ago
That is pretty cool.
12 years ago
Cool? It's more than just cool GH - we both remember when not everybody had a TV, colour was a novelty and 3d images had to be viewed under special lighting conditions in galleries you paid to get into.Images like this are something I used to dream about when I used to read Sf under the covers with a torch.
Oh I know (I remember seeing someone with one of those early "colorization panels you put over the tv screen to make the sky blue, and the grass green, etc, ...we've come a long way, for sure (I still prefer the picture I get from my radio over the tv though....but that is a personal thing). Images like this are something I used to dream about when I used to read Sf under the covers with a torch.You and I both. I still remember the first "laser produced hologram" that could be walked around. I immediately got an Edmund Scientific catalog and looked up what kind of lasers I would need etc. One laser alone would have cost me over $800 back then, but then, that is closer to what $2,000 is worth now. Thankfully the prices have dropped since then. *sigh* In this case "pretty cool" translates as "pretty amazing". ;-)
Yeah, that's Debevec's stuff. Many of the spiffy new CG techniques in use in the last 10 years can be traced directly or indirectly back to Debevec's work. Normal maps, image based illumination, High dynamic range photography and compositing . . .
That last shot of the guy walking is shooting literally Gigabytes of data per second through the projector.
Yeah, it's astounding how much data would be required to change your POV like this (and how many cameras would be needed for non-CGI images, etc.)
Is it something we'll see in the future, media-wise? No doubt.
In this case, they used 3 cameras as the subject was rotated. Thus the current system can only capture cyclical motion.
That would work for the static objects, but not for the 'running' man, which changes temporally from each angle. I'm thinking a bank of cameras, like the 'matrix' F/X...
They interpolate some data to fill in angles between cameras, too.
For this particular example, the running man video, only three cameras were used with a real person running. They were rotated around a subject performing the cyclical action. Thus, they could do it with less than a billion cameras. The particular limitation of this method (with three cameras) is that only cyclical action can be recorded, otherwise when the camera array is moved, the motion does not match up with recordings from other angles.For non-cyclical motion you would need a ton of cameras surrounding the subject.
The real limitation is that the objective (viewing camera or person) would have to move in sync with the original cameras. Otherwise I don't 'buy' that accurate perspective can be maintained...
Here is the PDF file for the capture method. They create a synthetic camera array, by rotating their three high speed cameras over 36 walk cycles. They assume that each walk cycle is the same, and frames from different cycles are aligned with target morphing by an optical flow algorithm.Then on playback with the "holographic" system, they pumped out about 5000 fps to maintain the illusion. That's one image per every 1.25 degree of viewing angle, i.e. automatic horozontal parrallax. The first walk around the head demonstrates this. They have to use viewer tracking to provide vertical parrallax as shown in the second look at the 3d head.
Sheesh - now *there*'s a hack! :-DOf course, the killing app for 3D systems is not so much 360 degree parallax - which is what this rotating mirror is great at, but depth perception. And for that, you don't need an expensive and incredibly bulky "bullet time" camera array. Something with a much smaller angle will do. Heck, you can probably rig a single camera with rotating mirrors or lenses instead of a camera array.This actually reminds me of a high-falutin' desktop 3D display I saw some 15 years or so ago. It used a membrane that was vibrating back and forth by a few centimeters (this was in Europe, so it was vibrating in centimeters), on which the image was projected. Looking back on it, I bet their main problem must have been sound insulation - since they were essentially putting people in front of a 40 cm speaker...
Well, with a rotating mirror you could get the effect of the subject rotating, but not movement without texture mapping and doing a CG camera move.
That's the key--it's just a repeating animation (like a gif anim), sampled over multiple cycles. So, doesn't fit my definition of a true temporal animation.
Thanks for the links...
ooh I want one! someone post an instructable of how to do it.
Just need a high speed projector, a holographic mirror spinning at 5000 cycles/sec and a huge dataset and computer to feed the projector.
3-d Technology Hall of Fame A bunch of different 3-d technologies.