Background Removal Tool 🎨

An AI-powered online background removal tool built with FastAPI, PyTorch, and the U²-Net deep learning model. Features a beautiful, modern UI with drag-and-drop functionality and real-time preview.

✨ Features

• 🎯 High-Quality Background Removal using U²-Net AI model
• 🖼️ Drag & Drop image upload
• 👁️ Live Preview with before/after comparison
• 🎨 Background Options (transparent, white, black, custom colors)
• 📥 One-Click Download with PNG transparency
• 📱 Responsive Design works on all devices
• ⚡ Fast Processing optimized for performance

🏗️ Architecture

The project uses a modern, modular architecture with clear separation of concerns:

removals-background/
├── app/                    # Main application package
│   ├── main.py           # FastAPI application
│   ├── config.py         # Configuration management
│   ├── api/              # API routes
│   ├── core/             # Core logic (model management)
│   ├── services/         # Business services
│   ├── models/           # AI model architectures
│   └── utils/            # Utilities
├── client/               # Frontend
│   ├── index.html
│   ├── styles.css
│   └── script.js
├── models/               # Model weights directory
├── requirements.txt      # Python dependencies
└── README.md            # This file

For detailed architecture documentation, see ARCHITECTURE.md.

🚀 Quick Start

Prerequisites

• Python 3.8 or higher
• pip (Python package manager)
• 4GB+ RAM recommended
• GPU (optional, but faster)

Installation

1. Clone or navigate to the project directory:

cd remove-bg

2. Create a virtual environment (recommended):

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Download the U²-Net model weights:

The model will be automatically downloaded on first run, or you can download it manually:

# Automatic download
python app/utils/download_model.py

# Or manually create models directory and download
mkdir -p models
# Download from: https://drive.google.com/uc?id=1ao1ovG1Qtx4b7EoskHXmi2E9rp5CHLcZ
# Save as: models/u2net.pth

Running the Application

1. Start the backend server:

Option 1: Using the startup script (recommended)

# Windows
start.bat

# Linux/Mac
chmod +x start.sh
./start.sh

Option 2: Using main.py

python main.py

Option 3: Using uvicorn directly

python -m uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

2. Start the frontend (React + Vite):

cd client
npm install  # or bun install
npm run dev  # or bun run dev

The frontend will be available at http://localhost:5173 (or the port shown in terminal)

Note: Make sure to set VITE_API_URL in client/.env if your backend is running on a different port.

3. Try it out!
   • Drag and drop an image or click to upload
   • Wait for processing (usually 3-10 seconds)
   • View the result and try different backgrounds
   • Download your image with transparent background

🔧 Configuration

API Endpoint

The frontend is configured to connect to http://localhost:8000. If you change the backend port, update the VITE_API_URL in client/.env:

VITE_API_URL=http://localhost:8000

Model Options

You can configure the model through environment variables or .env file:

# Use lighter U²-Net-P model for faster processing
MODEL_TYPE=u2netp

# Or use full U²-Net for better quality
MODEL_TYPE=u2net

# Force CPU usage
DEVICE=cpu

# Enable/disable advanced features
USE_MULTI_SCALE=true      # Better quality, slower
MASK_SMOOTHING=true        # Smoother edges
EDGE_REFINEMENT=true       # Better edge detection

See .env.example for all configuration options.

📝 API Documentation

Once the server is running, visit:

• Swagger UI: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc

Endpoints

GET /

Health check endpoint

{
  "message": "Background Removal API is running",
  "model_loaded": true,
  "device": "cpu"
}

POST /remove-background

Remove background from an image

• Input: multipart/form-data with image file
• Output: PNG image with transparent background

Example with curl:

curl -X POST "http://localhost:8000/remove-background" \
  -H "accept: image/png" \
  -F "file=@your-image.jpg" \
  --output result.png

🎨 Customization

Frontend Styling

The frontend uses React + Vite + Tailwind CSS. To customize:

• Edit client/tailwind.config.js for theme customization
• Edit client/src/assets/style/index.css for custom styles
• Components are in client/src/App.tsx

Backend Processing

Edit service/app.py to:

• Adjust image preprocessing
• Change model parameters
• Add new endpoints

🐛 Troubleshooting

Backend won't start

• Make sure all dependencies are installed: pip install -r requirements.txt
• Check if port 8000 is already in use
• Verify Python version: python --version (should be 3.8+)

Model not loading

• Ensure model file exists at service/models/u2net.pth
• Check file size (should be ~176MB)
• Verify download completed successfully

Frontend can't connect to backend

• Check that backend is running on port 8000
• Verify CORS is enabled in service/app.py
• Check browser console for errors

Processing is slow

• Use GPU if available (CUDA-enabled PyTorch)
• Switch to U²-Net-P (lighter model)
• Reduce input image size

Out of memory errors

• Reduce image size before processing
• Use U²-Net-P instead of U²-Net
• Close other applications

🚀 Deployment

Deploy Backend (Options)

1. Docker (recommended)
2. Heroku with buildpack
3. AWS EC2 or Google Cloud
4. DigitalOcean App Platform

Deploy Frontend

1. GitHub Pages (static hosting)
2. Netlify or Vercel
3. AWS S3 + CloudFront

Note: Update the API_URL in script.js to your deployed backend URL.

📊 Model Information

U²-Net (U Square Net)

• Paper: U²-Net: Going Deeper with Nested U-Structure for Salient Object Detection
• Architecture: Nested U-structure for accurate segmentation
• Size: ~176MB (full model), ~4.7MB (portable)
• Performance: High-quality results on portraits and objects

🤝 Contributing

Contributions are welcome! Feel free to:

• Report bugs
• Suggest features
• Submit pull requests

📄 License

This project is open source and available under the MIT License.

🙏 Acknowledgments

• U²-Net model by Xuebin Qin et al.
• FastAPI framework
• PyTorch team

📞 Support

If you encounter issues:

1. Check the troubleshooting section
2. Review the API documentation
3. Check browser console and server logs
4. Ensure all dependencies are correctly installed

---

Built with ❤️ using FastAPI, PyTorch, and U²-Net