In one embodiment, a method includes accessing a video; detecting one or more objects in one or more frames of the video; identifying one or more of the detected objects; determining a relevance score for each of the one or more of the identified objects with respect to a user the video is to be presented to; selecting one or more frames of the video based on the determined relevance scores for the identified objects in the frames; and providing for presentation to the user one or more of the selected frames of the video.