Jenji — Space Station Safety Object Detection

Jenji is a real-time object detection application built to detect critical safety objects on a space station using YOLOv11. It leverages Python, Flask-SocketIO, OpenCV, and React to stream and annotate live webcam feeds, with a WebView desktop launcher for easy deployment.

Jenji AI Glimpse

Jenji AI Scores

Jenji AI Project Structure

Jenji/
├─ .vscode/                    # VS Code configuration
├─ assets/                     # Project assets (images, logos)
├─ dataset/                    # YOLO dataset
│  └─ data.yaml               # YOLO dataset config
├─ ENV_SETUP/                  # Files for setting up conda environment
├─ runs/                       # Training outputs (weights, logs)
├─ runs_test/                  # Examine folder independent of project
├─ src/                        # Source code
│  ├─ training/
│  │  ├─ train.py             # Training script
│  │  ├─ eval.py              # Evaluation script
│  │  └─ config.yaml          # YOLO config
│  ├─ api/
│  │  ├─ app.py               # Optional REST API for image upload
│  │  ├─ config.py            # Configuration settings
│  │  ├─ detection_service.py # Detection service logic
│  │  ├─ detection_visualizer.py # Visualization utilities
│  │  ├─ image_processor.py   # Image processing utilities
│  │  ├─ socket_handlers.py   # WebSocket handlers
│  │  ├─ live_app.py          # Flask + SocketIO for live detection
│  │  └─ model_loader.py      # YOLO model loader
│  ├─ utils/
│  │  ├─ metrics.py           # Evaluation metrics
│  │  └─ viz.py               # Visualization utilities
│  ├─ frontend/                # React frontend
│  │  ├─ package.json
│  │  └─ src/
│  │     ├─ App.tsx
│  │     └─ index.tsx
│  └─ launcher/
│     └─ launch_app.py        # WebView-based desktop launcher
├─ .dockerignore               # Docker ignore file
├─ .gitignore                  # Git ignore file
├─ compose.yaml                # Docker Compose configuration
├─ Dockerfile                  # Docker container configuration
├─ download_dataset.ps1        # PowerShell script to download dataset
├─ environment.yaml            # Conda environment specification
├─ Makefile                    # Build automation
├─ plot.py                     # Plotting utilities
├─ README.Docker.md            # Docker-specific documentation
├─ README.md                   # Main documentation
└─ yolo11n.pt                  # Pre-trained YOLO11 model weights

🎯 Hackathon Objective

Detect 7 critical space station safety objects under varying conditions:

• OxygenTank
• NitrogenTank
• FirstAidBox
• FireAlarm
• SafetySwitchPanel
• EmergencyPhone
• FireExtinguisher

Key Goals:

• Train a robust YOLO model on synthetic data from Duality AI's Falcon simulator
• Evaluate model performance using [email protected], Precision, Recall, and Confusion Matrices
• (Bonus) Create a desktop or mobile app to use the trained model live

🚀 Quick Start

./download_dataset.ps1
make train
make run 

Frontend

make install-client
make client

Backend

make backend # after dataset downloaded and model train

⚙️ Setup Instructions

1. Clone Repository

git clone https://github.com/nee-hit476/Jenji.git
cd Jenji

2. Set up Python Environment

# Using conda
conda env create -f environment.yml
conda activate jenji

Dependencies include: torch, opencv-python, flask, flask-socketio, numpy, webview, socketio-client, etc

3. Download Dataset

Ensure dataset/ contains:

• train/, val/, test/ folders
• YOLO-compatible .txt labels
• data.yaml describing class names and dataset paths
• Run download_dataset.ps1

4. Train YOLO Model

cd src/training
python train.py --cfg config.yaml --data ../dataset/data.yaml --epochs 50 --batch-size 16

Outputs:

• Trained weights in runs/yolov11_experiment_x/weights/
• Logs and metrics in runs/yolov11_experiment_x/

5. Evaluate Model Performance

python eval.py --weights ../runs/yolov11_experiment_x/weights/best.pt --data ../dataset/data.yaml

Generates:

• [email protected] scores
• Confusion matrices
• Precision/Recall metrics

6. Run Flask-SocketIO Live Detection

cd src
python launcher/launch_app.py

• Opens WebView desktop window pointing to frontend http://localhost:5173
• Streams webcam frames to Flask server
• Returns annotated frames in real-time

7. Frontend Setup (React)

cd src/frontend
npm install
npm run dev

• Ensure the frontend dev server is running on http://localhost:5173
• Displays live annotated video frames from YOLO

8. Using the App

1. Allow camera permissions
2. Wait for the live feed
3. YOLO detects objects and overlays bounding boxes:
   • Green box = detected object
   • Label = class name + confidence

🐛 Debugging and Issues Faced

Issue	Fix
Webcam not working	Check browser permissions; close other apps using webcam
WebSocket errors	Ensure backend is running on http://localhost:8000
No detections	Check MODEL_PATH in model_loader.py; verify weights exist
Slow training	Reduce batch size; monitor GPU usage with nvidia-smi

---

📝 License

This project was created for a hackathon. Please check the repository for license details.

🤝 Contributing

Contributions, issues, and feature requests are welcome!

---

Made with ❤️ for Space Station Safety