Sunday, April 1, 2012

// // 3 comments

Jürgen Schmidhuber Talk on Ted, Creative Machines and the Omega Point | Computer Vision, 3D-reconstruction from 2D

Juergen Schmidhuber - TEDxLausanne - When creative machines overtake man



Funny as usual. :)

One critics I've got on the computers who will reach human brain computing power - brain power calculations are unreliable and not really comparable, and ironically brain computing power is specialized and probably some of the operations are redundant or unnecessary., A computer having power to simulate a brain is much more flexible and powerful. I've discussed on these issues on the blog and the AGI list. (Should do the digest...)

Computer vision problems of 3D-reconstruction from images and image and video processing, done with GPUs give some suggestions on supposed required computing power in digital terms, but human brain puts a lot of "empty cycles" before reaching to such capabilities, even while having actuators which are usually lacking yet in machines - if the agent is capable to manipulate the position of the object, the "cameras" and the focus predictably and slowly, as babies can do with their eyes, heads and body, it's far easier to detect and scale perspective laws etc.

And it takes many months to get to 3D-vision and to increase resolution and develop 3D-reconstruction in the human brain. That adds ~86400 fold per day and 31,536,000 "cycles" per year.

What computing power is needed?

I don't think you need millions of the most powerful GPUs and CPUs at the moment to beat human vision, we'll beat it pretty soon, a lot of the higher level intelligence in my estimation is very low at its complexity (behavior, decision making, language at the grammar/vocabulary levels) and would need a tiny amount of MIPS, FLOPS and memory. It's the lowest levels which require vast computing power - 3D-reconstruction from 2D one or many static or motion camera sources, transformations, rotations, trajectories computations etc., and those problems are practically being solved and implemented.

See for example:

Marc Pollefeys - ZURICH.MINDS



My estimations supercomputers were fast enough since many years, it's just the algorithms which are lacking yet.


*Thanks to Sasho for the link to the Juergen's presentation!


Words, Tags: Computer vision, 3D-reconstruction, 3D-graphics, computational creativity, Juergen Schmidhuber, Jürgen, Talks, Artificial General Intelligence, AGI, Futurology, Brain, Analysis, GPU, GPGPU, CPUs, computing power, supercomputers, 3D-models

3 коментара:

Alex Kamburov said...

Which reminds me of ProFORMA: Probabilistic Feature-based On-line Rapid Model Acquisition http://www.youtube.com/watch?v=vEOmzjImsVc

Todor "Tosh" Arnaudov - Twenkid said...

Thanks for the link, I remember this publication also. :)

This one is also related, but it's more specialized - the robot folding towels. :)

" Autonomously folding a pile of 5 previously-unseen towels"

http://www.youtube.com/watch?v=gy5g33S0Gzo

Generally the message is that 3D-reconstruction is already possible, sometimes in real time for simple objects, even not with the top-notch technology. Towels-folding is slow, but I rather believe their algorithm/actuators are not good enough yet, than that the problem is so complex to compute.

Todor "Tosh" Arnaudov - Twenkid said...

Clickable link: Robot folding towels