For the second time in its four-year history, the Large Scale Visual Recognition Challenge saw dramatic improvements in the quality of machine-vision technology.
The challenge was launched in 2010 by scientists from Stanford, Princeton, and Columbia universities with the goal of advancing the quality of image-recognition technology. The challenge uses the open source Imagenet database of more than 14 million images, which have been tagged and identified by humans.
Visual systems this year were tested in six categories based on their ability to detect objects, locate specific items in an object, and classify those images. The winners included Google, Adobe Systems, the National University of Singapore, and the Chinese Academy of Sciences.
Olga Russakovsky, the lead organizer of this year's contest, said the accuracy rate of this year's contestants nearly doubled from 22.5 percent to 43.9 percent, while error rates fell by almost half from 11.7 percent to 6.6 percent.
Fei-Fei Li, director of Stanford's Artificial Intelligence Laboratory, called this year's results "historic."
The improvement is attributed largely to expanded computing power that enabled the use of an approach known as a convolutional neural network by most of the competitors. The technique has existed since the late 1990s, but sufficient computing power to make use of it was not cost effective until recently.
From The New York Times
View Full Article
Abstracts Copyright © 2014 Information Inc., Bethesda, Maryland, USA
No entries found