Visual Dynamics: Probabilistic Future Frame Synthesis via Cross Convolutional Networks