How It Works

Upload Your Photo

Simply drag and drop or click to upload any photo you're curious about. Our system works with various image formats and sizes.

Advanced AI Analysis

Our state-of-the-art AI system analyzes your photo by:

  • Identifying key landmarks and architectural features
  • Recognizing natural landscapes and environments
  • Analyzing visual patterns specific to different regions
  • Processing environmental and cultural context clues

Location Prediction

Based on the analysis, we predict the most likely location where your photo was taken. For each prediction, we provide:

  • Precise geographic coordinates
  • Interactive map visualization
  • Confidence level of our prediction
  • Nearby landmarks and points of interest

Privacy & Security

We take your privacy seriously. All uploaded photos are processed securely and are not stored permanently on our servers. Your data is only used for location prediction and is automatically deleted after processing.

Get Started

Ready to discover where a photo was taken? Head back to our home page and upload your first image!

Technical Background

Our technology is based on the research paper "Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation" by Nicolas Dufour, David Picard, Vicky Kalogeiton, and Loic Landrieu.

Research Overview

Global visual geolocation predicts where an image was captured. Traditional methods struggle with ambiguous images (like beaches that could be from many locations) but excel with specific landmarks (like the Eiffel Tower). This research introduces a novel approach to handle this ambiguity.

Key Technologies

  • Diffusion Models: Progressive refinement of location predictions through noise reduction
  • Flow Matching: Efficient mapping from noisy to accurate coordinates
  • Riemannian Flow Matching: Earth surface-aware calculations for improved accuracy
  • Probabilistic Prediction: Multiple possible locations for ambiguous images

Performance

The system has been validated on major datasets (OpenStreetView-5M, YFCC100M, iNat21) and demonstrates:

  • State-of-the-art geolocation accuracy
  • Superior handling of ambiguous locations
  • Scalable performance on large datasets
  • Robust and interpretable results
Inspired by Nicolas Dufour et al. (2024). Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation. https://arxiv.org/abs/2412.06781