Google Neural Network – It’s easy to identify the location of a photo with an obvious landmark, but what about those photos taken inside a house, or an image of a food or a pet? It’s surprisingly easy for humans to identify the location, but similarly difficult for a machine. However, Google has made it possible thanks to its new artificial intelligence system that is outperforming humans in identifying the origins of images.
Google just unveiled a new system named PlaNet, a neural network that depends on image recognition technology to locate photos. Using meaningful visual hints such as building styles, languages, grooming of people and plants, the code matches those against a database of 126 million geotagged photos organized into 26,000 grids. For example, it could tell a photo was taken in Brazil based on the Portuguese signs and lush vegetation. It is also able to guess the locations of any indoor photos by using other images from the album as a starting point.
“PlaNet is able to localize 3.6 percent of the images at street-level accuracy and 10.1 percent at city-level accuracy,” said the research team.
Albeit PlaNet is not a foolproof system, but it already outperforms even the best globetrotters. To compare its accuracy to that of a human, the researchers matched their program against 10 well-travelled people in the game named Geoguessr, that provides random street-view photo and requires players to identify where they think the photo was taken, and there were 50 rounds in total.
“PlaNet won 28 of the 50 rounds with a median localization error of 1131.7 km, while the median human localization error was 2320.75 km,” the researchers said.
The program depends on internet images, so you can’t expect it to be 100 percent accurate. However, it’s really an impressive work to bring the superhuman neural network on your smartphone.
The researchers trained the neural network using 29.7 million public photos from Google+. The neural network relies on clues and features from photos it has already seen to help identify the most likely whereabouts of a new image.
The program has some limitations. Because it depends on internet images, PlaNet is at a disadvantage when confronted with rural countrysides and other rarely photographed locales. The team also left out large swaths of the Earth, including oceans and the polar caps.
Tobias Weyland, the lead author on the project, noted that supplementing internet photos with satellite images resolved some of these weaknesses. PlaNet also focuses on landscapes and other factors besides landmarks, making it more accurate at identifying non-city images than other programs.
One more piece of good news for users interested in identifying locations on the go: PlaNet only needs 377 MB, so it could fit on smartphones.