car_core/README.md

# Car Damage Inspector (Hugging Face-ready)

An open-source, one-file Gradio app that:
1) Classifies car damage type with a transformer classifier.
2) Detects damage regions & severity (Light/Moderate/Severe) with YOLOv8.

**Models used**
- `beingamit99/car_damage_detection` (image classification)
- `nezahatkorkmaz/car-damage-level-detection-yolov8` (YOLOv8 detector)

## Quickstart (local)

```bash
python -m venv .venv && source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt
python app.py
```

Open the Gradio link shown in the terminal and upload a car photo.

## Deploy to Hugging Face Spaces

1. Create a new **Space** (Python).  
2. Upload these three files:
   - `app.py`
   - `requirements.txt`
   - `README.md`
3. The Space will auto-build, download models on first run, and serve the UI.

## Notes
- First inference will download model weights; allow a little time.
- Gate logic is heuristic: if the top classification score < 0.5 (or model predicts an explicit *no-damage* class), the app returns "No visible damage".
- Replace/extend with your own gate (e.g., a dedicated binary model) if desired.
- Outputs:
  - **JSON** with top-3 labels and YOLO detections,
  - **Overlay image** with bounding boxes,
  - **Raw YOLO JSON** (for debugging/integration).