BayFlood offers a novel two-stage approach for detecting urban flooding from street-level imagery without requiring labeled training data.

• Uses zero-shot classification to identify potential flood images
• Applies Bayesian modeling to reduce false positives and improve accuracy
• Effectively detects flooding events from widely available street scene datasets
• Provides a scalable solution for urban infrastructure monitoring

This engineering breakthrough enables cities to leverage existing camera networks for detecting infrastructure problems, potentially reducing response times and damage costs during flooding events.

Bayesian Modeling of Zero-Shot Classifications for Urban Flood Detection