If you now let your imagination flow, you can predict tons of awesome uses for different disciplines.
Basically, it’s pure 3D real life data mining made possible.
For example, you can imagine that for a RPG, you can submit 10 Go of crowd videos to your 3D data mining system, and then get a database of different behaviours depending on different human interactions.
And then inject this nectar into your own RPG crowd system, in no time. Monthes of work saved.
It’s just an example. Another one coming to my mind, more creepy, is behavior prediction :
A surveillance cam system would analyze in real time how any passing human would react.
It would determine if an agression is being made, or even is about to being made, with an estimation by matching other 3D data of agressions …
It doesn’t seem very important as it is in the video, but it can really change the world if done well.
It looks truly amazing! But what exactly do you mean by that matrix “I know kungfu” clip, saying that it’s made more possible now? Can you expand bit more?
For the actual times, this means that biomechanical implants like this one, or even future artificial muscles will be able to have the precise motions for advanced movements like martial arts, dances, etc.
Just by digging and refining thousands of gigs of videos.
In a creepy future, robots will just be able to perfectly reproduce every single advanced kind of human movement on earth, by data mining video archives.
I’m not quite convinced this is so easy… it looks like they have keyframes which maybe they have to draw by hand or something, and then the video system figures out the in-between parts by tracking the movement in the video? Still sounds like quite a bit of work.
I’m not quite convinced this is so easy… it looks like they have keyframes which maybe they have to draw by hand or something, and then the video system figures out the in-between parts by tracking the movement in the video? Still sounds like quite a bit of work.
Judging from the low quality of the motions in that video, animators have nothing to worry about. It’s neat that it’s done from a single camera setup, but it’s quality is pretty far behind regular mocap systems.
I’d also second the previous comment on how the keyframes are being established in the first place - it indeed looks like they’re manually drawing in the keyframes and then their solver is just filling in the tweens… like a 2d-to-3d rotoscoping system (which would be useful even to animators).
Another advantage over mocap is the inherent smoothness of the interpolation. But it would be interesting to see the fail cases (Im guessing there are plenty).
sweet…don’t you need bullet trackers though as the ones needed in a major production? is this procedural or realtime?
I’m still confused as to how it doesn’t use trackers. I need that script.
It’s neat that it’s done from a single camera setup, but it’s quality is pretty far behind regular mocap systems.