The focus of this project is to decide the classification basis for our proposed yoga learning application. Our two proposed methods are a one-step end-to-end deep learning model and a two-step joint coordinate classification model. The one-step model would take in images and output a pose label – all feature engineering would be done by the hidden layers of the neural net. The two-step model would first use a deep learning model to locate the coordinate points of the joints in the image. Then, those joint coordinates would be fed into a classification model which would output the pose label.
