Droid210/FleetVision
0
Fleet-Vision =============
Fleet-Vision is a vehicle inspection and dispatch demo. It combines:
- Model A: vehicle body classification
- Model B: damage detection with heatmaps
- OpenRouteService: route distance and ETA lookup
- FastAPI: upload and dispatch endpoint
Setup
git lfs pull
pip install -r requirements.txtOr with Poetry:
poetry installRun
Start the API:
uvicorn api.main:app --reloadOpen the app in your browser and upload the five required images plus a destination.
Key files
- api/main.py: FastAPI app and
/request-rideendpoint - langchain_orchestrator.py: model and routing workflow
- body_classifier.py: Model A wrapper
- damage_detector.py: Model B wrapper
- utils/routing_service.py: ORS routing and geocoding
Model Documentation
For detailed Model A and Model B architectures, hyperparameters, evaluation metrics, and training guides, see models/README.md.
Notes
- Model weights are stored with Git LFS under
weights/. - Dataset folders live under
data/.
CI/CD: GitHub -> Hugging Face (Docker Space)
This repository includes a Docker-based CI/CD flow:
- CI workflow:
.github/workflows/ci.yml - Builds the Docker image on GitHub Actions.
- Runs a smoke test against
GET /health. - Deploy workflow:
.github/workflows/deploy-hf-space.yml - Syncs repository content to a Hugging Face Space repo.
- Hugging Face then builds and runs the app from
Dockerfile.
Required GitHub secrets
Set these in your GitHub repository settings -> Secrets and variables -> Actions:
HF_TOKEN: Hugging Face user access token with write permission to Spaces.HF_SPACE_REPO: Space identifier in the formusername/space-name.
Required Hugging Face Space secret
Set this in your Hugging Face Space settings -> Variables and secrets:
ORS_API_KEY: OpenRouteService API key used by routing service.
Deploy behavior
- Any push to
maintriggers CI and deploy workflows. - Deploy copies files to the Space repository and pushes a commit.
- If there are no file changes, deploy exits without creating a commit.
