TSP: Temporally-Sensitive Pretraining of Video Encoders for Localization Tasks
KAUST DepartmentComputer Science Program
Computer, Electrical and Mathematical Science and Engineering (CEMSE) Division
Visual Computing Center (VCC)
Electrical and Computer Engineering Program
KAUST Grant NumberOSR-CRG2017-3405
Permanent link to this recordhttp://hdl.handle.net/10754/666126
MetadataShow full item record
AbstractDue to the large memory footprint of untrimmed videos, current state-of-the-art video localization methods operate atop precomputed video clip features. These features are extracted from video encoders typically trained for trimmed action classification tasks, making such features not necessarily suitable for temporal localization. In this work, we propose a novel supervised pretraining paradigm for clip features that not only trains to classify activities but also considers background clips and global video information to improve temporal sensitivity. Extensive experiments show that using features trained with our novel pretraining strategy significantly improves the performance of recent state-of-the-art methods on three tasks: Temporal Action Localization, Action Proposal Generation, and Dense Video Captioning. We also show that our pretraining approach is effective across three encoder architectures and two pretraining datasets. We believe video feature encoding is an important building block for localization algorithms, and extracting temporally-sensitive features should be of paramount importance in building more accurate models. The code and pretrained models are available on our project website.
CitationAlwassel, H., Giancola, S., & Ghanem, B. (2021). TSP: Temporally-Sensitive Pretraining of Video Encoders for Localization Tasks. 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). doi:10.1109/iccvw54120.2021.00356
SponsorsWe present TSP, a novel temporally-sensitive supervised pretraining for video encoders, which not only trains to classify actions, but also considers background clips and global information to gain temporal sensitivity. We show that TSP features improve SOTA methods on the TAL, Proposals, and Dense-Captioning tasks. We argue TSP features can be preferred over other features to build more accurate models. Acknowledgments. This work is supported the King Ab-dullah University of Science and Technology (KAUST) Office of Sponsored Research (OSR) under Award No. OSR-CRG2017-3405.
Conference/Event name18th IEEE/CVF International Conference on Computer Vision Workshops, ICCVW 2021