+Massachusetts Institute of Technology (MIT)
researchers have developed a probabilistic programming language, dubbed Picture
, that can be used to infer the 3D shape of an object from 2D images with as little as 50 lines of code. Instead of requiring photo-realistic models which must be matched to input images, Picture
can be used to compare hypothesized scenes to observations, using a hierarchy of more abstract image representations to determine the correct representation for an image.
This method of “inverse graphics”, deducing the likeliest 3D model from 2D visual information, incorporates previous machine learning research and readjusts probabilities based on available training data. The researchers believe that this method of probabilistic programming will alleviate the re-writing of code when tackling different computer-vision problems. Read the full paper at http://goo.gl/WKkXZ2