Exploring Dimensional Collapse in Self-Supervised Video Representation Learning.