Google has created a computerized neural network that can identify the location of images with better accuracy than the most well-traveled of people, according to Gizmodo.
People can use many clues to place the location of a picture, like landmarks, backgrounds and architecture. But this new technology, called PlaNet, actually uses the pixels in a picture to establish where it was taken.
Tobias Weyand and his team have worked out a way of teaching a "deep-learning" machine how to use the pixels in a picture to map out location.
Using more than 26,000 squares, Weyand divided the world into a grid. Cities and busy areas had finer boxes on the grid, while open vast areas had larger boxes. The team then created a database with more than 100 million images complete with location, according to Technology Review. The end result was a neural network capable of determining a photo's origin with impressive accuracy.
"PlaNet is able to localize 3.6 percent of the images at street-level accuracy and 10.1 percent at city-level accuracy," said Weyand and his team. It can establish country of origin with an accuracy of about 28 percent and continent origin at just under 50 percent, according to GSM Arena.
The PlaNet system was put to the ultimate test when the machine was paired off against well traveled people, John Henry style.
"In total, PlaNet won 28 of the 50 rounds..." said Weyand. "[This] small-scale experiment shows that PlaNet reaches superhuman performance at the task of geolocating street view scenes."
