Google PlaNet trained to identify locations where photos were taken based on visual cues
Google PlaNet trained to identify locations where photos were taken based

Computer specialist Tobias Weyand and his colleagues at Google have come up with a deep-learning platform codenamed PlaNet. They have trained the system to figure out locations where pictures have been captured on the basis of visual cues. The project is interesting and it has showcased amazingly high degree of accuracy in detecting places where images were taken. The system is improving with time, said Google research team member Weyand.

The specialists at Google began by dividing up the globe into a grid, leaving aside the oceans and polar region. MIT Technology Review reported that thereafter, they formed a database for PlaNet, containing 126 million geo-located pictures taken from the Internet.

PlaNet is an artificial neural network, thus it can learn. So the experts taught the network how to recognize a picture’s location on the grid only with the help of information present in the pixels.

For testing the accuracy of PlaNet, Weyand and his team fed it 2.3 million geo-tagged Flickr pictures. Then, PlaNet lessened 48% of them to the right continent, 28.4% to the right country, 10.1% to the right city, and 3.6% to the real street.

With further improvement, PlaNet can exactly locate every single random picture on a map, however, it has to consider many other factors and pixels in pictures. In some cases, the system makes wrong guess about location.

Though the findings might not seem all that great initially, they became extraordinary when the tech giant team pitted their machine opposed to 10 smart, well-traveled humans. The machine managed to win over 50% of the rounds and had better accuracy.

The team wrote in their abstract regarding the system, “PlaNet outperforms previous approaches and even attains superhuman levels of accuracy in some cases”.

According to the team, PlaNet doesn’t need a lot of memory, either. MIT Technology Review reported that the model uses just 377 MB, which means it could be easily loaded on a smartphone.

PlaNet has distinct abilities, and it could end up being like a Shazam for picture locations. A user can run, Carmen Sandiego, but can’t stay out of sight of the Google machine.

Check out More news from Telecom Sector :: Pharmaceutical Sector :: Auto Sector :: Infrastructure :: Real Estate