Vehicle Detection using YOLO

Vehicle Detection application built with YOLOv11, FastAPI, and Gradio.
It provides an interactive web interface and REST API for running inference on images using a pretrained model.
It can predict: BUS, CAR, MICROBUS, MOTORBIKE, PICKUP-VAN, and TRUCK

---

Project Structure

ai_monk/
├── data/                          # raw training data (not required for inference)
├── runs/detect/vehicle_detection/ # trained YOLO model and logs
│   └── weights/
│       ├── best.pt                # pretrained weights (used for inference)
│       └── last.pt
├── src/
|   └── __init__.py                           
│   └── config.py                  # training utilities, configs
|   └── data_split.py
├── test_image/                    # sample test images
├── main.py                        # FastAPI + Gradio app for inference
├── train.py                       # training entrypoint
├── test_detection.py               # testing / evaluation script
├── requirements.txt
└── Dockerfile

---

Quick Start (Docker)

You can run this project without installing Python or dependencies — using Docker.

1 Clone the repository

git clone https://github.com/Amriteshwork/ai_monk.git
cd ai_monk

2 Build the Docker image

docker build -t vehicle-detector .

3 Run the container

Run the container and expose port 8000:

docker run -p 8000:8000 vehicle-detector

Now open your browser at: http://localhost:8000

You’ll see the Gradio interface for uploading images and viewing detection results.

---

Optional: Persistent outputs

To save the detected images on your host machine:

docker run \
  -p 8000:8000 \
  -v $(pwd)/inference_outputs:/app/inference_outputs \
  vehicle-detector

Detected images will appear in your local inference_outputs/ folder.

---

Optional: Custom model weights

If you want to run inference with your own trained weights, mount them and override the environment variable:

docker run \
  -p 8000:8000 \
  -v $(pwd)/my_weights:/app/weights \
  -e MODEL_WEIGHTS_PATH=/app/weights/best.pt \
  vehicle-detector

---

Development Setup (without Docker)

If you prefer running locally:

# Create a virtual environment
python3 -m venv .venv
source .venv/bin/activate   # On Windows use: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Run the app
python main.py

Then open http://localhost:8000

---

Configuration

The configuration is defined in src/config.py:

MODEL_WEIGHTS_PATH = Path(os.getenv(
    "MODEL_WEIGHTS_PATH",
    f"{BASE_DIR}/runs/detect/vehicle_detection/weights/best.pt"
))

• By default, it loads the pretrained model from runs/detect/vehicle_detection/weights/best.pt
• You can override this path using the environment variable MODEL_WEIGHTS_PATH

---

API Endpoint

Once running, you can also use the FastAPI endpoint:

POST /detect/
Upload an image file:

curl -X POST "http://localhost:8000/detect/" \
  -F "file=@test_image/test_car_3.jpg"

👨‍💻 Author

Amritesh Work
GitHub: @Amriteshwork

---