Communications of the ACM
Graphics in Reverse
Probabilistic programming does in 50 lines of code what used to take thousands.
Credit: MIT News
Programs of less than 50 lines written in a probabilistic programming language are just as effective as conventional systems with thousands of lines of code for completing some standard computer-vision tasks, according to researchers at the Massachusetts Institute of Technology (MIT).
"The whole hope is to write very flexible models, both generative and discriminative models, as short probabilistic code, and then not do anything else," says MIT graduate student Tejas Kulkarni.
The MIT researchers considered four computer-vision problems, each of which involves inferring the three-dimensional (3D) shape of an object from two-dimensional (2D) information. The technique is known as inverse graphics, an idea that was first developed with the original artificial intelligence research.
One of the tasks the researchers focused on is constructing a 3D model of a human face from 2D images. The new program describes the face as containing two symmetrically distributed objects--the eyes--with two more centrally positioned objects beneath them, representing the nose and mouth. If the program is fed enough examples of 2D images and their corresponding 3D models, it can create the eyes, nose, and mouth on its own.
The researchers found the new technique has an error rate 50-percent to 80-percent lower than conventional programs.
From MIT News
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found