Sign Language Recognition

AI-powered sign language detection using deep learning and React.

---

Quick Start

Windows:

run.bat

Choose options from the menu to install, train, and run.

Manual Setup:

# 1. Install dependencies
pip install -r requirements.txt
cd frontend && npm install && cd ..

# 2. Train model (30-60 min)
python ML/train_v2.py

# 3. Start backend (Terminal 1)
python backend/main.py

# 4. Start frontend (Terminal 2)
cd frontend && npm run dev

URLs:

• Frontend: http://localhost:3000
• Backend: http://localhost:8000
• API Docs: http://localhost:8000/docs

---

Project Structure

SLI/
├── backend/              # FastAPI server
│   ├── main.py          # API endpoints
│   ├── model_v2.onnx    # AI model
│   └── class_labels.txt # 43 sign classes
├── frontend/            # React app
│   ├── src/            # Components
│   └── dist/           # Production build
├── ML/                 # Training scripts
│   ├── train_v2.py     # Main training
│   └── inference.py    # Test model
├── data/               # Training images (43 classes)
├── run.bat            # Main launcher
└── requirements.txt   # Python dependencies

---

Features

• Real-time webcam detection
• 43 sign language phrases
• Multi-language translation (9 languages)
• Detection history tracking
• 85-95% accuracy
• FastAPI backend with REST API
• React frontend with TailwindCSS

---

Tech Stack

Backend:

• Python, TensorFlow, FastAPI, ONNX Runtime

Frontend:

• React, Vite, TailwindCSS, Framer Motion

Model:

• EfficientNetB3 (transfer learning)
• Input: 300×300 RGB images
• Output: 43 classes
• Size: ~50MB

---

Troubleshooting

Port in use:

netstat -ano | findstr :8000
taskkill /PID <pid> /F

Model not found:

python ML/train_v2.py

Dependencies error:

pip install -r requirements.txt --force-reinstall

---

Resources

• API Docs: http://localhost:8000/docs
• TensorFlow: https://www.tensorflow.org
• FastAPI: https://fastapi.tiangolo.com

---