Comment by isaacfrond

Anybody know what model architectures they used?

The motion detection might be a convolutional network or an svm. The mixer model perhaps a classic neural network.