README.md

metadata

title: Person Classification Demo
emoji: 📱
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false

📱 MobileNetV2 Image Classification Demo

A lightweight, interactive image classification demo built with Hugging Face Transformers and Gradio. This demo replicates the clean interface style of popular Whisper demos but for computer vision classification tasks.

✨ Features

• Mobile-Optimized Model: Uses MobileNetV2, designed specifically for efficient mobile and edge deployment
• Interactive Web Interface: Clean, modern Gradio interface similar to Whisper demos
• Real-time Classification: Instant image classification with top-5 predictions
• 1000 ImageNet Classes: Recognizes a wide variety of objects, animals, vehicles, and scenes
• Confidence Scores: Shows prediction confidence as percentages
• Example Images: Pre-loaded example images for quick testing
• Responsive Design: Works seamlessly on desktop and mobile devices
• Lightweight: Only 3.4M parameters for fast inference

🚀 Quick Start

Option 1: Hugging Face Spaces (Recommended)

Deploy instantly to Hugging Face Spaces:

1. Create a new Space on Hugging Face Spaces
2. Upload these files: app.py, requirements.txt, README.md
3. Your demo will be live automatically!

Option 2: Local Development

# Clone this repository
git clone <your-repo-url>
cd mobilenetv2-classification-demo

# Install dependencies
pip install -r requirements.txt

# Run the demo
python app.py

The demo will be available at http://localhost:7860

🎯 How to Use

1. Upload Image: Click on the upload area or drag and drop an image
2. Get Results: Classification happens automatically, showing top-5 predictions
3. Try Examples: Use the example buttons to test with sample images
4. View Confidence: Each prediction shows a confidence percentage

📊 Model Information

• Model: google/mobilenet_v2_1.0_224
• Architecture: MobileNetV2 with 1.0 width multiplier
• Input Size: 224×224 pixels
• Parameters: 3.4 million (lightweight!)
• Classes: 1,000 ImageNet categories
• Optimization: Designed for mobile and edge devices

🔧 Technical Details

Dependencies

• PyTorch: Deep learning framework
• Transformers: Hugging Face model library
• Gradio: Web interface framework
• Pillow: Image processing
• NumPy: Numerical computing

Model Architecture

MobileNetV2 uses:

• Depthwise Separable Convolutions: Reduces computational cost
• Inverted Residuals: Efficient feature extraction
• Linear Bottlenecks: Maintains representational power
• ReLU6 Activation: Optimized for mobile hardware

🌟 Key Advantages

vs. Heavy Models (ResNet, EfficientNet)

• ✅ Faster inference (optimized for mobile)
• ✅ Smaller memory footprint
• ✅ Better battery efficiency
• ✅ Edge deployment ready

vs. Other Demos

• ✅ Whisper-style interface (familiar UX)
• ✅ Auto-classification (no manual buttons needed)
• ✅ Clean, modern design
• ✅ Mobile-responsive

📱 Mobile Deployment

This model is specifically designed for mobile deployment:

# Example mobile optimization
model = MobileNetV2ForImageClassification.from_pretrained(
    "google/mobilenet_v2_1.0_224",
    torch_dtype=torch.float16,  # Half precision
    low_cpu_mem_usage=True      # Memory optimization
)

🎨 Customization

Adding New Examples

example_urls = {
    "Your Category": "https://your-image-url.com/image.jpg",
    # Add more examples here
}

Adjusting UI Theme

theme = gr.themes.Soft()  # Options: Soft, Default, Monochrome

Changing Model

MODEL_NAME = "google/mobilenet_v2_1.4_224"  # Larger variant
MODEL_NAME = "google/mobilenet_v2_0.75_224"  # Smaller variant

🔄 Image Processing Pipeline

1. Input: User uploads image (any format/size)
2. Preprocessing: Resize to 224×224, normalize
3. Inference: MobileNetV2 forward pass
4. Postprocessing: Apply softmax, get top-5
5. Output: Formatted predictions with confidence

🚀 Performance

• Inference Speed: ~50ms on CPU, ~10ms on GPU
• Memory Usage: ~200MB RAM
• Model Size: ~14MB
• Throughput: 20+ images/second on modern hardware

📚 Example Classes

The model can classify 1,000 ImageNet categories including:

• Animals: Dogs, cats, birds, wildlife
• Vehicles: Cars, trucks, motorcycles, aircraft
• Objects: Furniture, electronics, tools
• Food: Fruits, vegetables, dishes
• Nature: Plants, landscapes, weather

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Make your improvements
4. Submit a pull request

📄 License

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

🙏 Acknowledgments

• Google Research: For MobileNetV2 architecture
• Hugging Face: For the Transformers library and model hosting
• Gradio Team: For the amazing web interface framework
• ImageNet: For the comprehensive dataset

🔗 Links

• MobileNetV2 Paper
• Hugging Face Model
• Gradio Documentation
• Live Demo

---

Built with ❤️ using Hugging Face Transformers and Gradio