In this paper, we propose a dynamic graph modeling approach to learn spatial-temporal representations for video summarization. Most existing video summarization methods extract image-level features ...