Install the required libraries for this can be installed via
pip install -r ./requirements.txt
Please note that pytorch can be installed via the pytorch website to allow training via GPU.
The dataset in its full form is found in ./data. The dataset can also be found and downloaded online here.
There are three possible models to train in the dataset (SVM, KNN, and CNN). These are found in the ./models file as Python files. To train these models, run their corresponding files in ./models. These files will split the dataset into a train and test set, train the model, save the model in ./models/trained_models, and test the model. The test accuracy for the model, as well as the confusion matrix for the model, will be outputted once training is complete.
Please note that for the KNN, validation has shown that k=153 is the best possible k value. ./models/knn.py contains the code to validate the value of K, but, by default, it only trains a model at k=153. Refer to lines 101-102 to enable validation, but know that this will increase training time to 20-40 minutes.
Running ./live_feed.py will attempt to create a connection to the camera, identify faces in frame, and will assign an emotion to the face. Live_feed utilizes the most accurate model, the CNN.