Real-Time Emotion Detection using AffectNet

This project implements a real-time emotion detection system using a deep learning model trained on the AffectNet dataset. The model is capable of recognizing eight emotions: Angry, Disgust, Fear, Happiness, Neutral, Sadness, Surprise, and Contempt.

Features

• Real-time facial emotion recognition using OpenCV.
• Deep learning model trained on the AffectNet dataset.
• Supports multiple face detection in a single frame.
• Pretrained model (.h5 format) for quick deployment.

Dataset

This model is trained using the AffectNet dataset, a large-scale facial expression dataset containing annotated images of different human emotions. The dataset was preprocessed by resizing images to 96x96 pixels and normalizing pixel values.

Dataset link: AffectNet on Kaggle

Installation

To set up the project, first install the required dependencies:

pip install -r requirements.txt

Training the Model

To train the model from scratch, run the following script:

python train.py

This will:

1. Load and preprocess images from the AffectNet dataset.
2. Train a deep learning model with Convolutional Neural Networks (CNNs).
3. Save the trained model as emotiondetector.h5.

Running Real-Time Emotion Detection

To run real-time emotion detection using a webcam, execute:

python main.py

Model Architecture

The model consists of multiple convolutional layers, batch normalization, dropout, and fully connected layers. It uses the Softmax activation function in the final layer to classify emotions.

Dependencies

The following libraries are required for the project:

• TensorFlow/Keras
• OpenCV
• NumPy
• Pandas
• Scikit-learn

Author

This project was developed as part of an emotion recognition system using deep learning techniques.

---

requirements.txt

tensorflow
keras
opencv-python
numpy
pandas
scikit-learn
matplotlib
tqdm

Ensure you have Python 3.7+ installed before installing dependencies.