Oops! Predicting Unintentional Action in Video

Did this person intend for this action to happen, or was it an accident? In this paper, we introduce a large in-the-wild video dataset of unintentional action. Our dataset, which we have collected by downloading "fail" videos from the web, contains over twenty thousand clips, and they span a diverse number of activities and scenes. Using this dataset, we study a variety of visual clues for learning to predict intentionality in video.