Overview
The majority of research in action understanding focuses on designing methods to encode a few seconds of short, trimmed clips and classify these with single action labels.
Such methods, however, are rarely applicable for temporally localizing and/or classifying actions from longer streams of video.
In this tutorial, we would like to focus on research on understanding actions long videos up to tens of minutes.
Compared to action recognition from trimmed video clips long video understanding tasks pose more challenges due to the long span of videos and complex temporal relations between occurring actions. Such challenges include: “What are the actions and when do these actions happen in the long video sequences?” Our main focus for this tutorial is two tasks that aim to find human actions in videos, i.e., Temporal Action Detection/Localization (TAD/L) and Temporal Action Segmentation (TAS).
Speakers
Angela Yao Assistant Professor, National University of Singapore
Junsong Yuan Professor, SUNY Buffalo
Hilde Kuehne Assistant Professor, Goethe University
Ehsan Elhamifar Associate Professor, Northeastern University
Yogesh S Rawat Assistant Professor, University of Central Florida
Schedule
Opening Remarks (9:00 - 9:10am, Organizers)
Invited Speaker: Angela Yao (9:10 - 9:45am) [slide]
Invited Speaker: Junsong Yuan (9:45 - 10:20am) [video]
Coffee Break (10:20 - 10:40am)
Invited Speaker: Hilde Kuehne (10:40 - 11:15am) [video]
Invited Speaker: Ehsan Elhamifar (11:15 - 11:50am) [video]
Invited Speaker: Yogesh S Rawat (11:50 - 12:25pm) [video] [slide]
Conclusion (12:25 - 12:30pm, All)
Resources
Work Compliation
Survey
Temporal Action Segmentation: An Analysis of Modern Techniques, 2022. Guodong Ding, Fadime Sener and Angela Yao. [PDF]
Weakly-supervised Temporal Action Localization: A Survey, 2022. AbdulRahman Baraka and Mohd Halim Mohd Noor. [PDF]
A Survey on Temporal Action Localization, 2020. Huifen Xia and Yongzhao Zhan [PDF]
Organizers
Angela Yao Assistant Professor, National University of Singapore
Junsong Yuan Professor, SUNY Buffalo
Fadime Sener Research Scientist, Reality Labs, Meta
Guodong Ding Research Fellow, National University of Singapore