Facial Emotion Detection System

A deep learning-based facial emotion detection system that can recognize emotions in real-time and recommend music based on the detected emotion.

Features

• Real-time facial emotion detection using webcam
• Emotion classification into 5 categories: Angry, Fear, Happy, Sad, Neutral
• Music recommendation system based on detected emotion
• User-friendly Gradio web interface
• Support for diverse music recommendations (English, K-pop, and Burmese songs)

Project Structure

Facial_Emotion_Detection_2/
├── data/
│   └── FER2013/           # Dataset directory
│       ├── train/         # Training dataset
│       └── test/          # Testing dataset
├── haarcascades/
│   └── haarcascade_frontalface_default.xml  # Face detection model
├── model/
│   ├── emotion_cnn.py     # CNN model architecture
│   └── emotion_cnn_best.pth  # Trained model weights
├── gradio_app.py          # Gradio web interface
├── real_time_detection.py # Real-time detection script
├── train.py              # Model training script
└── requirements.txt      # Project dependencies

Technologies Used

• Python 3.x
• PyTorch for deep learning
• OpenCV for image processing
• Gradio for web interface
• Matplotlib for visualization

Model Architecture

The emotion detection model uses a CNN architecture with residual blocks:

• Initial convolution layer
• 3 Residual blocks with increasing channels (64, 128, 256)
• Dropout layers for regularization
• Batch normalization for stable training
• Fully connected layers for final classification

Installation

1. Clone the repository
2. Install dependencies:

pip install -r requirements.txt

Usage

Web Interface

Run the Gradio web application:

python gradio_app.py

Real-time Detection

For real-time webcam detection:

python real_time_detection.py

Training

To train the model:

python train.py

Music Recommendation System

The system includes a curated list of songs for each emotion:

• Diverse selection of English, K-pop, and Burmese songs
• Emotion-appropriate playlists
• Direct YouTube links for easy access

Dataset

The project uses the FER2013 dataset for training and testing, which includes facial expressions categorized into different emotions.

Requirements

Major dependencies include:

• torch==2.6.0
• opencv-python==4.11.0.86
• numpy==2.2.3
• pandas==2.2.3
• matplotlib==3.10.1

For a complete list of dependencies, see requirements.txt.

Contributing

Feel free to open issues or submit pull requests for improvements.

License

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