期刊
出版社
IEEE COMPUTER SOC
DOI: 10.1109/CVPR46437.2021.00012
关键词
-
资金
- National Key Research and Development Program [2018YFB0804204]
- National Defense Basic Scientific Research Program [JCKY2020903B002]
- Strategic Priority Research Program of Chinese Academy of Sciences [XDC02050500]
- National Nature Science Foundation of China [62022078, 62021001, 62071122]
- Open Project Program of the National Laboratory of Pattern Recognition (NLPR) [202000019]
- Youth Innovation Promotion Association CAS [2018166]
The proposed Uncertainty Guided Collaborative Training (UGCT) strategy effectively improves the performance of attention based methods for weakly supervised temporal action detection by generating pseudo labels online and mitigating noise in the generated labels. Experimental results show a significant performance improvement of more than 4% for all three methods on the THUMOS14 dataset.
Weakly supervised temporal action detection aims to localize temporal boundaries of actions and identify their categories simultaneously with only video-level category labels during training. Among existing methods, attention based methods have achieved superior performance by separating action and non-action segments. However, without the segment-level ground-truth supervision, the quality of the attention weight hinders the performance of these methods. To alleviate this problem, we propose a novel Uncertainty Guided Collaborative Training (UGCT) strategy, which mainly includes two key designs: (1) The first design is an online pseudo label generation module, in which the RGB and FLOW streams work collaboratively to learn from each other. (2) The second design is an uncertainty aware learning module, which can mitigate the noise in the generated pseudo labels. These two designs work together to promote the model performance effectively and efficiently by imposing pseudo label supervision on attention weight learning. Experimental results on three state-of-the-art attention based methods demonstrate that the proposed training strategy can significantly improve the performance of these methods, e.g., more than 4% for all three methods in terms of mAP@IoU=0.5 on the THUMOS14 dataset.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据