Shocking! Now You Can Track Location With Images
Graduate students at Stanford University have developed an application called Predicting Image Geolocations (PIGEON) that utilizes generative AI to detect a specific location from a street view or image
PIGEON demonstrates impressive accuracy in predicting the country depicted, achieving a success rate of 92%. Moreover, it can pinpoint a location within 25 kilometers of the target in over 40% of its predictions, showcasing its effectiveness
PIGEON's performance was tested against GeoGuessr, a popular game where users guess the location of a photo from Google Street View. The AI ranked within the top 0.01% of GeoGuessr players, highlighting its exceptional capabilities
In a series of six matches streamed online, PIGEON outperformed one of the world's best professional GeoGuessr players, Trevor Rainbolt, garnering over 1.7 million views
The students leveraged CLIP, a neural network developed by OpenAI, to connect text and images. The model was trained on a dataset that linked visual categories with their corresponding names
The PIGEON project underscores the delicate balance between the positive applications of AI in image geolocalization and the potential risks, emphasizing the importance of responsible development and deployment of such technologies