On the Compactness, Efficiency, and Representation of 3D Convolutional Networks: Brain Parcellation as a Pretext Task (2017-06-25T00:00:00.000000Z)