This repository contains the code and documentation for a Facial Emotion Classification project. The project explores the classification of human emotions using a unique Convolutional Neural Network (CNN) architecture. Departing from pre-trained models, the CNN is designed for enhanced precision and insights, achieving a 64.02% accuracy on unseen data.
- Introduction
- Dataset
- Methods and Theory
- Model Architecture
- Implementation
- Results
- Usage
- Contributing
- License
- Acknowledgements
Emotions are a crucial aspect of human expression, and this project aims to classify seven different emotions based on facial expressions using a bespoke CNN architecture. Unlike traditional transfer learning approaches, the model is constructed from the ground up, allowing for fine-tuning and experimentation.
The project utilizes the FER 2013 dataset, a freely available collection of grayscale images with seven emotion classes: anger, happiness, neutrality, surprise, fear, sadness, and disgust. Meticulous dataset partitioning and data augmentation techniques were implemented to enhance the model's exposure to real-world scenarios.
The project employs various deep learning methodologies, including convolutional layers, pooling layers, batch normalization, ReLU activation, dropout layers, and fully connected layers. The model's architecture is experimented with different batch sizes and epochs to achieve optimal performance.
The CNN architecture comprises 38 layers, organized into convolutional blocks A and B, and a fully connected block. The model is designed to accurately classify facial expressions while addressing challenges in emotion classification.
The complete model is implemented in Python using the Keras API for the TensorFlow library. Google Cloud GPU-enabled Colab platform is utilized to accelerate the training process, reducing the training time to 30 minutes for 60 epochs.
The model achieves a 64.02% accuracy on unseen data, as evaluated through accuracy metrics, loss curves, confusion matrix, and heat maps. The results highlight the model's strengths in classifying certain emotions and areas for potential improvement.
To use the code, follow these steps:
- Clone the repository:
git clone https://github.com/your-username/facial-emotion-classification.git - Install dependencies:
pip install -r requirements.txt - Run the main script:
python main.py
Contributions are welcome! Please follow the contribution guidelines when submitting pull requests.
This project is licensed under the MIT License.
- Keras API
- Google Cloud Colab
- FER 2013 Dataset
- References (See the References section in the project report for the complete list)