Stanford University has been showing off rather cool 3-D modeling that takes a 2-D image and creates a navigable 3-D image.

According to the site, for each small homogeneous patch in the image, a Markov Random Field (MRF) is used to infer a set of "plane parameters" that capture both the 3-D location and 3-D orientation of the patch. The site said that the MRF, when trained, creates image depth cues as well as the relationships among different parts of the image.

It gives a much richer experience in the 3-D flythroughs created using image-based rendering, even for scenes with significant non-vertical structure. Stanford has set up a Web page that allowed people to upload a picture of their own and see how it worked. Unfortunately, it was spotted by Slashdot and interest was so high that the server crashed.

More here.