Building Detector

AI-powered building detection from satellite imagery using SAM2.

Building Detector is a portfolio project by a Geoinformation student. The web app lets users select an area on a map, place guide points on buildings, and run SAM2-based detection with footprint regularization. Results can be exported as standard GeoJSON or in an OpenStreetMap-compatible format.

Features

• Point-guided building detection from satellite imagery
• SAM2-based segmentation with footprint regularization
• Satellite image download through leafmap
• Dual export formats: standard GeoJSON and OSM-compatible GeoJSON
• Support for either a local SAM2 backend or a Colab-hosted backend

Architecture

The project uses a simple two-part architecture:

• Flask web app for the browser UI, imagery download, and export flows
• SAM2 backend running either locally or through Colab for detection processing

UML diagrams and architecture notes are available in the UML/ directory.

Quickstart

1. Clone the repository.
2. Install the main application dependencies:

   pip install -r requirements.txt

3. Start the Flask app:

   python main.py

4. Configure either the local backend or the Colab backend before running detection.

For detailed backend setup, see local_backend/README.md.

Tech Stack

• Python
• Flask
• SAM2 / segment-geospatial
• Leafmap
• Leaflet.js
• Rasterio
• GeoPandas

License

This project is licensed under the MIT License. See LICENSE.