Overview¶
OpenFloodAI is an open-source project for low-cost river flood detection support. Use a fixed camera near a river, run checks close to the camera, and create clear records when something may need human review.
OpenFloodAI is not a finished public warning system. Early code must not send public warnings or make emergency decisions by itself. Current output means "please review this evidence," not "there is a confirmed flood."
Which one do I want?¶
- I'm a developer or contributor — see the Developer Overview and Developer Quickstart.
- I'm a hydrologist, disaster-management staff, or evaluating this as a field tool — see the Overview and Quickstart written for you, no coding required.
What's here¶
-
Product
Requirements, quality checklists, and the ML readiness plan that decide what gets built next.
-
Architecture
How the pieces fit together: data contracts, reason codes, windowed video evidence, and validation input snapshots.
-
Validation & Research
Labeling guides, human-label comparisons, threshold tuning, and known validation limits.
-
Learning
Guided walkthroughs of a full end-to-end review and the core pipeline basics.
-
Packaging & Release
How the desktop app is built and released for Windows and macOS.
-
Decisions
Architecture Decision Records — significant technical choices and why they were made.
Current status¶
OpenFloodAI has a usable local validation MVP. The project is still a proof of concept, but contributors can prepare labelled examples, run validation, inspect scorecards and evidence, and compare recent local reports without cloud services.
OpenFloodAI still does not detect real floods, train ML models, send alerts, publish warnings, run live camera deployments, or replace local emergency decision-making. The Home UI is a local validation and review tool, not a production monitoring or fleet dashboard.
The current runtime uses Python, jsonschema, numpy, and OpenCV. Development
checks use pytest, Ruff, mypy, and MkDocs. See the dependency map.