In this work, we identify two key observations about spatiotemporal redundancy in videos: Temporal redundancy is not bound to fixed spatial locations. Semantically consistent elements in videos often ...