I’ve seen examples of multiple pictures being stitched together into panoramas and even three dimensional models, but this work from Carnegie Mellon is particularly impressive in that it builds models from a single photograph. Be sure and check out the little movies on this page.
Somehow this makes me think of a cybernetic version of the bloodhound that can track down a criminal based only on an old handkerchief. Pretty soon, we’ll be able to take a scrap of a movie poster from Titanic, say something with just a fragment of the boat and Leonardo DiCaprio’s nose on it, and we’ll wave it at the computer and say “Come on boy! There you go! It’s melodramatic historical fiction. Now go! Go get ‘im little feller!” And after grinding and chugging for a few minutes, the computer will churn out the entire film.