Search Street View for Anything!

Back in 2024, all text in NYC, performed large-scale character recognition across every Google Maps Street View image in New York, producing a search engine that lets users find any word captured in the city’s streetscape.

Now, Sean Hardesty Lewis at Cornell Tech has pushed the idea further - using vision-language AI models to build an interactive map that lets users search not just for words, but for virtually anything visible in the city, from objects and materials to styles, atmospheres, and abstract visual patterns.

Searchable City allows you to search New York's Google Maps Street View imagery for just about anything. For example

https://searchable.city/?q=statue

returns Street View images of statues across the city, identifying and mapping each instance detected by the system.

Searchable City isn’t limited to clearly defined objects; it can also surface more diffuse, abstract, or atmospheric qualities. For example,

https://searchable.city/?q=secluded

reveals Street View scenes that feel tucked away or off the beaten track, highlighting quieter, less visible corners of the city.

What makes Searchable City so compelling is that it provides a new direction in map search. Traditional maps organise space through fixed labels - addresses, categories, and boundaries - whereas Searchable City reorganises the city through perception. It surfaces patterns that are usually invisible to spatial databases: allowing you to surface where scaffolding clusters, where flowers can be enjoyed, or which streets feel tidy, or neglected.

In doing so, the project hints at a different kind of urban interface - one where the city becomes searchable by meaning as well as by location.