In this project, we applied Convolution Neural Networks to predict emotions from the image of human faces. We used Kaggle Facial Emotion Recognition 2013 [11] dataset for training. The human accuracy on this dataset is 65%. We experimented with different models and hyper parameters to finally come up with a model that meets the human accuracy. Our best performing model also performs in the range of the top 4 models of the challenge. Further, we applied different visualisation techniques on this model.
Forked from https://github.com/mihaelacr/pydeeplearn
Borrowed the webcam live video capture from the Pydeeplearn repository
Added support for training CNNs to recognize facial expressions using Keras Deep Learning library