Emotion Analysis Web App

A real-time emotion detection web application that uses deep learning to identify seven basic emotions from webcam input.

Model Performance

The emotion detection model achieves the following performance metrics on the test set:

• Accuracy: 67.59%
• Loss: 0.9084

Training Plots

Confusion Matrix

The confusion matrix shows the model's performance across different emotion categories. The diagonal elements represent the number of correct predictions for each emotion class.

Features

• Real-time webcam capture and processing
• Emotion detection for seven basic emotions (Happy, Sad, Angry, Fear, Surprise, Disgust, Neutral)
• Web interface with live video feed
• Responsive UI with Bootstrap styling
• Face detection using OpenCV
• Deep learning model using TensorFlow

Screenshot

Project Structure

emotion_analysis/
├── app.py                 # Flask web application
├── download_model.py      # Script for model training
├── plot_training.py      # Script for generating training plots
├── requirements.txt      # Project dependencies
├── templates/           # HTML templates
│   └── index.html      # Main web interface
├── plots/              # Training plots and visualizations
│   └── confusion_matrix.png
└── FER-2013/          # Dataset directory
    ├── train/         # Training images
    └── test/          # Test images

Setup Instructions

1. Create a virtual environment:

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

2. Install dependencies:

pip install -r requirements.txt

3. Train the model:

python download_model.py

4. Generate training plots:

python plot_training.py

5. Run the application:

python app.py

6. Open a web browser and navigate to http://localhost:5000

Technologies Used

• Python 3.11
• TensorFlow
• OpenCV
• Flask
• HTML/CSS
• NumPy
• Matplotlib
• Seaborn

Model Architecture

The emotion detection model uses a Convolutional Neural Network (CNN) with the following architecture:

• Input layer: 48x48x1 (grayscale images)
• Multiple convolutional and pooling layers
• Dense layers with dropout for regularization
• Output layer: 7 units (one for each emotion)

Dataset

The model is trained on the FER-2013 dataset, which contains:

• 28,709 training images
• 7,178 test images
• 7 emotion categories
• 48x48 pixel grayscale images

License

This project is licensed under the MIT License - see the LICENSE file for details.