Long-Range Video Understanding

Angela Yao

National University of Singapore, SGP

Abstract

The hallmark task for video understanding is action recognition. Action recognition classifies pretrimmed video clips of a few seconds long with single semantic labels. But what about videos that run longer than a few seconds? This tutorial will introduce the challenges and state-of-art methods that work with long-range videos that span minutes, or even hours. We will give a special focus to procedural activities – activities that feature multiple actions in sequence, and discuss tasks such as recognition, anticipation and temporal segmentation.