Emotion Recognition using ResNet CNN Architecture

This project attempts to recognize user emotion using a convolutional neural network (CNN). The particular architecture used is a residual neural network based (ResNet).

The neural net can recognize 7 emotions with relatively high accuracy: (1) Anger, (2) Disgust, (3) Fear, (4) Happy, (5) Sad, (6) Surprise and (7) Neutral.

The dataset for training the neural net came from the Carrier and Courville Facial Expression Dataset hosted on Kaggle.

How to Run:

Emotion Recognition

(1) In order to get going quickly, run the face_tracking.py file and the program will begin to track your emotions via webcam.

Neural Net Training

(1) The neural net can be re-trained to obtain a different model via the emotion_recognition.py file.

The current model has an accuracy of ~94.8% on the test dataset.

Hardware Requirements

(1) Webcam, (2) Graphics card supporting Tensorflow

Note: Program has only been tested under Ubuntu 14.04 with an NVIDIA GTX 1070.

References:

(1) https://github.com/tflearn/tflearn/blob/master/examples/images/residual_network_cifar10.py

(2) https://www.kaggle.com/c/challenges-in-representation-learning-facial-expression-recognition-challenge