Improving Sequential Determinantal Point Processes for Supervised Video Summarization