[P] Built an open source tool to find the location of any street picture

Developer sparkyniner has released a free, accessible web demo for Netryx Astra V2, an open-source AI tool designed to solve the complex problem of image geolocation. The tool uses a sophisticated computer vision pipeline to analyze visual cues in a street-level photograph—such as architecture, street signs, vegetation, and road layouts—and pinpoint its geographic coordinates. Previously, users needed technical know-how to install and run the GitHub repository. The new demo removes that barrier, offering instant access to a pre-indexed 10km radius around New York City, allowing anyone to upload a photo and see where it was likely taken.

While the hosted web demo operates on a limited credit system to manage GPU inference costs, the fully open-source nature of the project empowers users to take control. Anyone can clone the GitHub repository, index their own city or region with custom street view imagery, and run unlimited searches locally. This dual approach—a user-friendly demo for experimentation and a self-hostable repo for power users—makes advanced geolocation AI accessible to a broad audience. The developer is actively soliciting feedback on failed searches to improve the model's accuracy and robustness across diverse urban environments.