🛠️ Tech Stack

🚀 Installation & Setup

Prerequisites

• Python 3.8+
• Node.js 16+
• FFmpeg

1 — FFmpeg Setup

• Download FFmpeg from Gyan Builds
• Extract the archive and add the bin folder to your system PATH environment variable

2 — Backend Setup (Django)

pip install pipenv          # First time only

cd backend
pipenv shell
pipenv install -r requirements.txt

python manage.py makemigrations label
python manage.py migrate label
python manage.py runserver

3 — Frontend Setup (React)

cd frontend
npm install
npm start

The application will be available at http://localhost:3000

📖 Usage

1. Register / Sign In — Create an account or log in to access your projects
2. Create a Project — Enter a name and description to initialise a new annotation project
3. Upload Media — Upload video files from the Manage Project section; extract frames at your desired FPS threshold
4. Annotate Frames
   • Manual: Draw, resize, and label bounding boxes directly on frames
   • Automatic: Run AI inference using a preloaded or custom YOLO model to auto-generate labels
5. Review & Export — Save bounding boxes for future editing; export in JSON, COCO, or YOLO format for downstream model training

🎬 Project Demonstration

📽️ View the full project demonstration here

🌟 Highlights

• Extensible AI Model Framework — Architected to support plug-and-play integration of a wide range of AI models, including Ultralytics-compatible models and certain other custom detection models, without requiring changes to the core annotation pipeline.
• Active Learning Loop — uncertainty sampling identifies low-confidence detections and routes them to human reviewers, creating an intelligent human-in-the-loop workflow rather than purely manual or purely automatic annotation
• Production-Ready Export — annotations export directly in COCO, YOLO, and JSON formats, compatible with major training frameworks (PyTorch, TensorFlow, Ultralytics)
• Multi-Version YOLO Support — benchmarked across YOLOv8, v9, and v10, allowing users to trade off between inference speed and detection accuracy depending on dataset requirements
• Peer-Reviewed Validation — system design and performance metrics independently validated through publication in IGI Global

🔮 Future Improvements

• Database scaling — migrate from SQLite to PostgreSQL for multi-user and production deployments
• Cloud storage integration — replace local file storage with AWS S3 or equivalent for scalable video and frame management
• Segmentation support — extend annotation beyond bounding boxes to polygon and pixel-level segmentation masks for semantic segmentation tasks
• Model performance dashboard — in-app visualisation of precision, recall, and mAP metrics per model to help users select the most appropriate model for their dataset
• Collaborative annotation — multi-annotator support with inter-annotator agreement scoring (Cohen's Kappa) for dataset quality control
• Auto-retraining pipeline — trigger model fine-tuning automatically as new verified annotations accumulate, closing the active learning loop end-to-end

👥 Team

📄 License

This project is licensed under the CC BY-NC 4.0.