In each frame, the upper left is the input; the upper right is the geometric tracking
illustrated by an overlayed yellow face mesh; the bottom is the recognized expression
visualized by training images.