There is more to images than their objective physical content: for example,advertisements are created to persuade a viewer to take a certain action. Wepropose the novel problem of automatic advertisement understanding. To enableresearch on this problem, we create two datasets: an image dataset of 64,832image ads, and a video dataset of 3,477 ads. Our data contains rich annotationsencompassing the topic and sentiment of the ads, questions and answersdescribing what actions the viewer is prompted to take and the reasoning thatthe ad presents to persuade the viewer ("What should I do according to this ad,and why should I do it?"), and symbolic references ads make (e.g. a dovesymbolizes peace). We also analyze the most common persuasive strategies adsuse, and the capabilities that computer vision systems should have tounderstand these strategies. We present baseline classification results forseveral prediction tasks, including automatically answering questions about themessages of the ads.
展开▼