In this project I used Convolutional Lstm neural network which is a combination of Convolutional and Long Short-Term memory neural networks allowing for spatial dependencies in data sequences like videos to be captured efficiently over time.
The dataset is here.
I implemented this neural network in both Pytorch and Keras frameworks.
I optimized the hyperparameters by training and testing the model over and over again, each time with different settings.
Due to GPU limitaions I restricted 11 classes of original dataset to only 'Biking' and 'Horse Riding' and our model achieved an accuracy of over 80 percent.