Face Mask Detection using Deep Learning

This project is an updated version of a team project I completed in 2021 (published in IEEE conference proceedings of ICACRS 2022), which previously only classified two categories: "with mask" and "without mask". This enhanced implementation now distinguishes a third category - "mask worn incorrectly", making it more comprehensive for real-world applications.

Features

• Real-time face mask detection through webcam
• Image upload for mask detection
• Responsive web interface
• High accuracy classification (97.83% on test data)

Model Performance

The model achieves excellent performance across all three classes:

Classification Report:
                        precision    recall  f1-score   support

mask_weared_incorrect       0.99      0.99      0.99       598
with_mask                   0.96      0.98      0.97       598
without_mask                0.99      0.96      0.98       598

accuracy                                        0.98      1794
macro avg                   0.98      0.98      0.98      1794
weighted avg                0.98      0.98      0.98      1794

Final validation accuracy: 97.83% Final validation loss: 0.0613

Technology Stack

• Python 3.9
• TensorFlow 2.12 with Metal support for Mac
• MobileNetV2 for the base convolutional neural network
• OpenCV for computer vision and face detection
• Flask for web application framework
• Bootstrap 5 for frontend styling
• JavaScript for interactive frontend functionality

System Requirements

• Recommended: macOS with Apple Silicon (M1/M2/M3)
• Python: 3.9.x
• At least 4GB RAM
• Webcam (for live detection)

Installation and Setup

1. Clone the repository

git clone https://github.com/421pvv/face-mask-detection-using-deep-learning.git
cd face-mask-detection-using-deep-learning

2. Create a Conda environment

# Create a new environment with Python 3.9
conda create -n mask-detection python=3.9
conda activate mask-detection

3. Install dependencies

pip install -r requirements.txt

Note for non-Mac users: If you're not using a Mac with Apple Silicon, you should replace tensorflow-macos and tensorflow-metal in the requirements.txt with standard tensorflow:

pip uninstall tensorflow-macos tensorflow-metal
pip install tensorflow==2.12.0

Project Structure

face-mask-detection/
├── data/
│   └── processed/           # Place your dataset here
│       └── mask_dataset/    # Dataset directory
│           ├── with_mask/
│           ├── without_mask/
│           └── mask_weared_incorrect/
├── models/
│   ├── face_detector/       # Face detection model files
│   └── face_mask_detection_model/ # Trained model for inference
├── notebooks/
│   └── model_training.ipynb # Model training notebook
├── static/
│   ├── css/
│   ├── js/
│   └── images/              # Store visualization images here
│       ├── training_history.png
│       └── confusion_matrix.png
├── templates/               # HTML templates
├── app.py                   # Flask application
├── detection_utils.py       # Utility functions for detection
└── requirements.txt

Dataset

The dataset used for training is from Kaggle: Face Mask Detection Dataset

Dataset Preparation

1. Download the dataset from the Kaggle link
2. Extract the dataset and organize it into three folders:
   • data/processed/mask_dataset/with_mask/
   • data/processed/mask_dataset/without_mask/
   • data/processed/mask_dataset/mask_weared_incorrect/

Model Training

If you want to train the model yourself:

1. Ensure you have the dataset organized as described above
2. Open the notebook notebooks/model_training.ipynb
3. Run the cells in sequence
4. The trained model will be saved to the models/ directory

The model uses MobileNetV2 architecture with transfer learning, which provides an excellent balance between accuracy and performance for deployment.

Running the Web Application

To start the web application:

python app.py

This will start the Flask development server, and you can access the application at http://localhost:3000.

Using the Application

1. Home Page: Provides options for live detection or image upload
2. Live Detection: Accesses your webcam for real-time mask detection
3. Image Upload: Allows you to upload an image for mask detection

Troubleshooting

Common Issues on macOS

• If you encounter webcam access issues, make sure to grant camera permissions to your terminal/IDE
• For M1/M2/M3 Macs, the application uses a Haar Cascade face detector as a fallback if the DNN detector fails

Common Issues on Other Platforms

• If you encounter CUDA/GPU errors on Windows/Linux, try disabling GPU acceleration by setting this environment variable:

  export CUDA_VISIBLE_DEVICES=-1  # Linux/Mac
  set CUDA_VISIBLE_DEVICES=-1     # Windows

Acknowledgements

• MobileNetV2 Paper
• Face Mask Detection Dataset
• Our Previous Work (2022) - Conference paper from ICACRS 2022