Simplified GeoGuessr: Visual Place Recognition from Google Street View Images

Description

Inspired by the worldwide popularity of the GeoGuessr game, which challenges players to guess locations from Google Street View images, we created a simplified version that predicted the city or country from a single street-view photo. This project was framed as a multi-class image classification task over a fixed set of cities. Each model focused on extracting structural or visual cues from the image using computer vision and vision–language modeling techniques. These models were combined into a single integrated system that determined the most probable cities based on the classification outputs. The final outcome was a working prototype capable of identifying the city and displaying its top predictions.

Demo Video