Siggraph 2014 kicks off next week, and everyone's getting their cool graphics demos ready. One of those demos, put together by folks at Carnegie Mellon and UC Berkeley, demonstrates a new image editing technique that allows subjects inside 2D photos to be manipulated in 3D. The results are impressive, to say the least:
The technique apparently involves "publicly available 3D models," which are used to "guide the completion of the geometry and appearance of the revealed areas of the object." Illumination is then estimated using information from the 2D source image. I'm not sure how effective the technique is with more complex and exotic objects, but it definitely looks to have potential.
You can read the team's full paper, watch a longer video of the technique in action, and download some OS X executables and source code on this page. (Thanks to TR fan Brian for the link.)