Fig. 6

The video summaries of ADL P_20.mp4 that is about 27 min long. The capturer does many things in the video: brushing his teeth, going to a laundry room, studying, watching TV, and washing dishes. The clustering-based method lost some important scenes, and the attention-based method tends to extract some keyframes when the camera wearer is moving, which makes a bad frame likely to be included in a consumer video’s summary