[Geeks are Sexy] technology news

Monday, July 10, 2006

Transforming a 2D Picture Into 3D: They Said It Couldn't Be Done

Transforming 2D pictures into 3D scenes has long been thought impossible, but lo and behold, researchers at Carnegie Mellon University have changed all that. They recently developed a new method that allows a computer to learn the geometric context of a 2D image, letting it automatically re-create it into 3D. Here's a little demo of what the technology can do.


  • That is very cool! Technology is truly amazing. I wonder what they will think of next.

    By Blogger Aaron, at 8:54 AM  

  • I thought of this before they did it. You have to make a few assumptions, and with only one phote to work from, the quality frankly sucks. What one would need to do to get a good 3D image is have multiple, shots, recognise features and vectors, extrapolate camera angles, and then unscale textures and background. With just one shot, you have to assume basically that most non vertical vectors are parallel to the ground, and that the camera is right-side-up. Then, you might use shadows to figure out where the sun is, then use that and shadows to figure other edges. I'd say that with only one photo, that's as good as it gets for the most part... I would reccomend using the sky as backdrop for the final image at least, to make the video look better. I've wanted for some time to build a camera and rack device that would take shots from pre-measured angles in a panoramic fasion expressly for the creation of 3d photos. But Google's on top of that with their new Google Earth San Fransisco (?) 3d campaign...

    By Blogger Ethan Anderson, at 10:14 PM  

Post a Comment

Links to this post:

Create a Link

<< Home