This paper proposes a novel approach and a new benchmark for video summarization, which focuses on user videos, which are raw videos containing a set of interesting events, and generates high-quality results, comparable to manual, human-created summaries.