Composable Augmentation Encoding for Video Representation Learning.