Deep learning

The objective is to implement a hybrid model that encompasses two deep learning algorithms suitable for emotion recognition and more precisely for processing multi-class and multi-label facial problems such as (CNN,GAN,RNN, LSTM) with an accuracy of 90% and more. DAISEE : multi-label video classification data set including 9068 video clips captured from 112 users will be used to recognize the user's affective states: Boredom, confusion, engagement and frustration, daisee dataset has four levels of labeling, namely - very low, low, high and very high for each of the affective states, and we are going to train the model on this dataset and then we are going to take the same daisee dataset and manually annotate it with new classes that are the six basic emotions (joy...), surprise, fear, anger, disgust and sadness) and will train our hybrid model on this new dataset and we will compare the experimentation result for each class and the accuracy of the result of each class of the classes treated whether for emotional states or basic emotions. For the quantity of the training data we can use just half of the dataset and the rest can be kept for the test.