Annotation Efficient Visual Recognition: From Semi-Supervised To Few-Shot Learning