End-to-end learning of visual representations from uncurated instructional videos

Conference item

Annotating videos is cumbersome, expensive and not scalable. Yet, many strong video models still rely on manually annotated data. With the recent introduction of the HowTo100M dataset, narrated videos now offer the possibility of learning video representations without manual supervision. In this work we propose a new learning approach, MIL-NCE, capable of addressing mis- alignments inherent in narrated videos. With this approach we are able to learn strong video representations from scratch, without the need for any manual annotation. We evaluate our representations on a wide range of four downstream tasks over eight datasets: action recognition (HMDB-51, UCF-101, Kinetics-700), text-to- video retrieval (YouCook2, MSR-VTT), action localization (YouTube-8M Segments, CrossTask) and action segmentation (COIN). Our method outperforms all published self-supervised approaches for these tasks as well as several fully supervised baselines.

Authors: Miech, A; Alayrac, J-B; Smaira, L; Laptev, I; Sivic, J

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: IEEE
• Host title: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
• Pages: 9876-9886
• Publication date: 2020-08-05
• Acceptance date: 2020-02-23
• Event title: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
• Event location: Online
• Event website: https://cvpr2020.thecvf.com/
• Event start date: 2020-06-14
• Event end date: 2020-06-19
• DOI: 10.1109/CVPR42600.2020.00990
• EISSN: 2575-7075
• ISSN: 1063-6919
• EISBN: 9781728171685
• ISBN: 9781728171692
• Language: English
• Keywords: videos; training; task analysis; visualization; kinetic theory; noise measurement; manuals