README.md

metadata

title: Car Detection
emoji: 🐢
colorFrom: gray
colorTo: yellow
sdk: docker
pinned: false
license: cc-by-nc-nd-4.0
short_description: car-counting

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Parking - Aerial Photo Car Detection

Detect and count cars in aerial/satellite photography using an RF-DETR object detection model, served via a FastAPI backend with a Preact frontend.

Architecture

training/     Model training, export, and inference scripts (RF-DETR + ONNX)
server/       FastAPI backend — serves ONNX model predictions
frontend/     Preact + Vite + TypeScript UI

Quick Start

Prerequisites

• Python 3.11+
• Node.js 18+
• Trained ONNX model at training/exported_models/inference_model.sim.onnx

Backend

pip install -r server/requirements.txt
uvicorn server.main:app --reload --port 8000

Frontend

cd frontend
npm install
npm run dev

Open http://localhost:5173, upload an aerial image, and view detection results.

API

POST /detect

Multipart form upload. Returns detected cars with annotated and heatmap images.

Parameter	Type	Default	Description
file	file	—	JPEG or PNG image
threshold	float	0.5	Confidence threshold 0–1

Response:

{
  "car_count": 47,
  "detections": [{"bbox": [x1, y1, x2, y2], "score": 0.97}, ...],
  "annotated_image": "data:image/jpeg;base64,...",
  "heatmap_image": "data:image/jpeg;base64,..."
}

GET /health

Returns {"status": "ok", "model_loaded": true}.

Deployment

The backend and frontend are deployed separately so the page loads instantly (static CDN) while the model runs server-side.

Backend — Hugging Face Spaces

1. Create a new Space at huggingface.co/new-space with Docker SDK
2. Clone the Space repo and copy in the required files:

   git clone https://huggingface.co/spaces/YOUR_USER/YOUR_SPACE
   cd YOUR_SPACE
   cp /path/to/parking/Dockerfile .
   cp -r /path/to/parking/server ./server
   mkdir -p training/exported_models
   cp /path/to/parking/training/exported_models/inference_model.sim.onnx ./training/exported_models/
   git add . && git commit -m "Initial deploy" && git push
   

3. The Space will build the Docker image and start serving at https://YOUR_USER-YOUR_SPACE.hf.space

Frontend — Vercel

1. Push the frontend/ directory to a GitHub repo (or use the full project repo)
2. Import the project at vercel.com/new
3. Set Root Directory to frontend
4. Add the environment variable:
   • VITE_API_URL = https://YOUR_USER-YOUR_SPACE.hf.space
5. Deploy — Vercel auto-detects Vite and builds it

Model

• Architecture: RF-DETR (medium)
• Single class: car
• Trained on aerial parking lot imagery in YOLO format
• Exported to ONNX with graph simplification
• Training scripts and data live under training/