A representation that decomposes scene dynamics into a compact set of motion bases, each as a temporal sequence of 6D twists.