Automated road damage assessment powered by Vision Transformers, built for field teams, deployable anywhere.
StreetSafe takes a road image and returns a structured damage report: defect type, severity level, and repair recommendation. The ML backbone is google/vit-base-patch16-224 served via a FastAPI backend. Field teams access it through a Flutter mobile app; internal teams use a Streamlit dashboard. The whole backend ships as a Docker container.
| Feature | Description |
|---|---|
| ViT Classification | Classifies road damage (Potholes, Cracks, Surface Wear) using google/vit-base-patch16-224 |
| Severity Mapping | Maps each defect class to High, Medium, or Low severity automatically |
| Repair Recommendations | Returns material requirements and estimated labor scope per defect type |
| Confidence Scoring | Includes match confidence in the JSON response for every prediction |
| Dual Frontend | Streamlit dashboard for internal use; Flutter app for field teams |
| Containerized Backend | Docker-ready — single command to spin up the API server |
ML Engine : Hugging Face Transformers, PyTorch, PIL
Backend : Python 3.13, FastAPI, Uvicorn
Web Dashboard : Streamlit
Mobile : Flutter (Dart)
DevOps : Docker, Git LFS