IEEE Computer Society Newsfeed

Subscribe RSS

« Back

Google Software More Accurately Describes Photos

Google has developed artificial-intelligence software capable of more accurately describing a photo scene than previous technologies, which could make searching for specific images easier or help the visually impaired better understand a photo. The technology uses two neural networks for processing. One focuses on image recognition and processing, the other handles natural-language processing and produces full English sentences to describe photos. In tests, the software, following training, was applied to images available in openly published datasets and asked to provide a description. Against an automated test used to benchmark computer-vision software, the Google technology—which is still in development—scored in the 60s. Humans taking the same test typically score in the 70s. (BBC)(WIRED UK)(MIT Technology Review)(Google Research)

Trackback URL: