Sequence Multi-Labeling: A Unified Video Annotation Scheme With Spatial and Temporal Context