Abstract

We introduce the novel problem of automatically gener-ating animated GIFs from video. GIFs are short loopingvideo with no sound, and a perfect combination between im-age and video that really capture our attention. GIFs tell astory, express emotion, turn events into humorous moments,and are the new wave of photojournalism. We pose thequestion: Can we automate the entirely manual and elabo-rate process of GIF creation by leveraging the plethora ofuser generated GIF content? We propose a Robust DeepRankNet that, given a video, generates a ranked list of itssegments according to their suitability as GIF. We train ourmodel to learn what visual content is often selected for GIFsby using over 100K user generated GIFs and their corresponding video sources. We effectively deal with the noisy web data by proposing a novel adaptive Huber loss in the ranking formulation. We show that our approach is robust to outliers and picks up several patterns that are frequently present in popular animated GIFs. On our new large-scalebenchmark dataset, we show the advantage of our approachover several state-of-the-art methods.