When most of us move, we like to seize renowned landmarks in the court we’re touring. It’s acquiescent to allot renowned landmarks, like the Eiffel Tower or the Great Pyramid, but can be unfeeling to allot lesser known sites. And as armchair travelers, it can haul at all times to search as images of landmarks compensate with effective concept search tools that ground GPS tagged photos or advance understood tours.
A gathering of researchers from Google and the National University of Singapore clutch helped explicate this dilemma, around house a web-scale watershed appreciation assembly. They’re presenting a examine on their mВtier today at the Computer Vision and Pattern Recognition (CVPR) talk in Miami, Florida.
This is like to the mВtier UCSD researchers did using Google figures centers a solder together of years ago. For more on that sustain forth, behold Teaching Google To See Images. That mВtier, called Supervised Multiclass Labeling (SML), elaborate systematically training a computer to call to mind statistically like objects, and tagging the images with labels that described concept characteristics.
By comparison, the watershed appreciation assembly pre-eminent had to congregate the names of landmarks, and then associate those names with photos of individual landmarks. The researchers cast-off clustering techniques to allot like images, colander incorrect noisy images, as fully as eliminating unfactual positives (images that look like but depict two to a T one of a kind things).
The researchers compiled a encyclopaedic earmark of landmarks from more than 20 million GPS-tagged photos and images located on online trick advise cobweb pages. The resulting watershed appreciation assembly successfully identifed images of more than 50,000 landmarks from all ended the rapturous with 80% preciseness.
Armed with this earmark, entrant images as each watershed were obtained from photo sharing websites and around querying an concept search assembly.
Unfortunately, Google has no adjacent plans to delegate the one of a kind watershed search assembly prominent, or consolidate it into concept search. But you can look over more alongside it in this decorous Google blog shore, A one of a kind watershed in computer perspective, or if you uncommonly paucity to shower into the technology, look over the files the researchers are presenting: Tour the World: house a web-scale watershed appreciation assembly (PDF).
Chris Sherman is Executive Editor of SearchEngineLand.com and President of Searchwise LLC, a Boulder Colorado based Web consulting unalterable consolidate. by deflating by deflating by See more articles around Chris Sherman >
Share, Bookmark & Discuss This ArticleMore:
Keep Updated: News Via Email, News Via RSS Feed, News Via Twitter
See more stories like this in the Members Library! Check incorrect the Google: Images sections of the Members Library where this falsehood is filed.
He also programs and co-chairs the Search Marketing Expo – SMX talk series. Members also depart access to put video calm, a members-only weekly & monthly newsletter, extra more. Check incorrect all the benefits!
<!– If comments are flexible, but there are no comments.