With the latest landmark update, Google improves its location recognition from photos

Google announced Friday, May 3, 2019 the update of its photographic database, which lists thousands of landscapes and monuments. Nourished with artificial intelligence, Landmarks now allows immediate recognition of these sites.

Identify a place for sure or recognize it on any image of a catalog: the new version of Google's Landmarks program, announced Friday, May 3, 2019, allows the Web giant to take off on the competition. The data collected is intended to encourage research in artificial intelligence and machine learning, before finding applications.

This second version allows, according to the American multinational, to refine the reading of images through a record compilation of various sites worldwide. Google has given amateur photographers the opportunity to feed their database. In the end connoisseurs of their daily environment, they could annotate precisely their shots. In parallel, the firm crossed these with the visuals free of right provided by Wikimedia Commons.

As a result, no less than five million photos are now cataloged by the Google service, double the previous version for some 200,000 sites around the world seven times more than in 2018! "This is the most complete dataset in the world," said Bingyi Cao and Tobias Weyand, software engineers at Google AI.

Because this is a tool for researchers. Since the launch of the first version of the program, the Web giant has also organized two competitions in which over 500 experts in machine learning participated. Faced with this craze, Google has not only decided to renew these events but also to publish Landmarks in open-source as part of a new home method, called "Detect-to-Retrieve". The latter would allow the analysis to be even more precise by isolating the characteristic elements of a site. "We hope that these improvements to our dataset will advance the state of the art in the field of visual recognition of monuments and landscapes," say the engineers on the Google AI blog.