Emotion Recognition Using Convolutional Neural Networks

Description

This project aims to implement emotion recognition using convolutional neural networks (CNNs), inspired by the architectures and methodologies described in "Facial Emotion Recognition: State of the Art Performance on FER2013" and "Facial Expression Recognition in the Wild via Deep Attentive Networks". The project utilizes TensorFlow and Keras for model development, with a focus on achieving high accuracy in detecting human emotions from facial images.

Features

• Utilizes a custom-built CNN architecture for emotion recognition.
• Employs Haar cascades for effective face detection in preprocessing.
• Implements data augmentation to enhance model robustness.
• Split data into training and validation sets for effective model evaluation.
• Can be extended to real-time emotion recognition with minimal modifications.

Model Architecture

The model architecture is based on the principles outlined in the referenced papers, adapted for emotion recognition. Key features include:

• Multiple convolutional layers for feature extraction.
• MaxPooling layers for spatial data reduction.
• Dropout layers to mitigate overfitting.
• Dense layers for classification, with a softmax activation function in the output layer to handle multiple emotion classes.

Training Dataset

The dataset used for training is the FER-2013. It contains images which are categorised to one of 7 categories being 0=Angry, 1=Disgust, 2=Fear, 3=Happy, 4=Sad, 5=Surprise, 6=Neutral. The features of the dataset:

• Training set of 28,709 examples in the training set, 3,589 in the testing set
• Faces are already centered and occupies roughly the same amount of space

Installation

To set up the project, follow these steps:

1. Clone the repository
   git clone https://github.com/jdl00/emotion_recognition

2. Install the required libraries
   pip install -r requirements.txt

Usage

1. To train the model run the train command

   # Environment variables
   
   # Folder to export the model to
   EXPORT MODEL_EXPORT_PATH='folder_to_export_model_run_to'
   
   # Dataset path to be used (classifiers are pulled from the dataset)
   EXPORT DATASET='dataset_path'
   
   # By default models are quantised and optimised, to disable this you can export
   EXPORT OPTIMISE=0 (to turn off optimisation)
   
   # Set the amount of epochs to train for
   EXPORT EPOCHS=200
   
   # Patience till training early stops
   EXPORT patience=5
   
   # Finally execute the train script
   'ENV_VARIABLES' python train.py
   

2. Perform recognition on camera input

   The realtime input to the model uses Pythons OpenCV library. You can specify which input by using the device_id

   # Environment variables:
   
   # Device ID of the camera
   EXPORT DEVICE_ID='device_id_of_camera'
   
   # Path to the exported model
   EXPORT MODEL_PATH='exported_model_path'
   
   # Finally execute the recognition script
   'ENV_VARIABLES' python recognition.py
   

Results

TODO: Complete this section.

Contributing

Contributions to this project are welcome. To contribute:

1. Fork the repository.
2. Create a new branch for your feature (git checkout -b feature/your_feature).
3. Commit your changes (git commit -m 'Add some feature').
4. Push to the branch (git push origin feature/your_feature).
5. Create a new Pull Request.

Note: Adjust these instructions based on how you wish others to contribute to your project.

License

MIT License