It's not often I'm sitting in front of the computer with my mouth hanging open, but the video below is literally jaw-dropping. Researchers Tao Chen, Zhe Zhu, Ariel Shamir, Shi-Min Hu and Daniel Cohen-Or have developed software called "3-Sweep," an insanely cool way to extract editable 3D data from a 2D image. Not a bunch of 2D images—we all know the technology where you walk around an object and fire off a dozen shots—just one image. Which means you no longer have to be there with a camera, but could conceivably pull a (relatively well-shot) 2D image from anywhere, and quickly create a model of it.
"Our approach combines the cognitive abilities of humans with the computational accuracy of the machine to solve this problem," writes the team. What you basically do is use your mouse to "sweep" lines and/or ellipses across the image, quickly teaching the software where the axes are. Look at how freaking easy this looks, and watch what they do with the telescope and lamppost:
We dig that they showed not only the successes, but also the failures of the software, to give you a realistic idea of what is possible. And yes, it seems to lend itself best to geometric objects with some degree of rotational symmetry, but this is still an extraordinary breakthrough.
For those of you in Hong Kong, a 3-Sweep demo is scheduled for the upcoming SIGGRAPH Asia.