SAM3 Vision AI for Smart City Entity Detection
Urban Planning & Government — Several US cities needed highly accurate detection of zoning features, property lines and urban infrastructure, along with the…
Client: US City Government. Sector: Urban Planning & Government.
The challenge: Several US cities needed highly accurate detection of zoning features, property lines and urban infrastructure, along with the ability to process video in real time
What we built: We fine-tuned SAM3 and Qwen multimodal models to detect entities in live video, identifying thousands of them across US cities. Models were trained on each city's own data, with edge compute deployment planned for the next real-time phase
Tech used: SAM3, Qwen, Fine-tuned Models, Edge Compute, Real-time Video AI, PyTorch.
- Accuracy of Detection: 98.7%
- Entities Found: 1000s in real-time
- Speed: Real-time video
- Cities Live: 12+