Abstract
This paper proposes motion prediction in single still images by learning itfrom a set of videos. The building assumption is that similar motion ischaracterized by similar appearance. The proposed method learns local motionpatterns given a specific appearance and adds the predicted motion in a numberof applications. This work (i) introduces a novel method to predict motion fromappearance in a single static image, (ii) to that end, extends of theStructured Random Forest with regression derived from first principles, and(iii) shows the value of adding motion predictions in different tasks such as:weak frame-proposals containing unexpected events, action recognition, motionsaliency. Illustrative results indicate that motion prediction is not onlyfeasible, but also provides valuable information for a number of applications.