.png){kind=icon}
返回
Deep snippet selective network for weakly supervised temporal action localization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.patcog.2020.107686.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Temporal action localization has been a hot topic in video analyzation. In this paper, we propose a novel method called deep snippet selective network (DSSN) to address two key problems in weak supervision for temporal action localization, which are separability and integrality. Specifically, we employ two eras-ing branches to ensure the integrality, which can force the network to select other complementary snippets by erasing the most discriminative snippets. It is worth mentioning that a ternary mask is utilized to provide erasing branches with a background prior to enhance the separability of the model. Besides, we design a background suppression branch to further reduce the effect of background snippets. Extensive experiments on dataset THUMOS'14 and ActivityNet show the effectiveness of our method. (c) 2020 Elsevier Ltd. All rights reserved.
Keyword:
Weak supervision
Temporal action localization
Erasing branches
Ternary mask
Background suppression branch
AI总结
对已上传原文的论文进行重点信息的提取，主要内容包括：简要概述、研究摘要、背景介绍、关键亮点、图文解析、展望与总结。
期刊
IF:
7.6
论文数:
1.3W
被引数:
4.5W
机构
引用论文
Learning principal orientations and residual descriptor for action recognition.png)
PATTERN RECOGNITION
PATTERN RECOGNITIONIF7.6
PREAI
分享收藏Learning motion representation for real-time spatio-temporal action localization实时时空动作定位的运动表示学习PATTERN RECOGNITION
PATTERN RECOGNITIONIF7.6
PREAI
分享收藏
分享收藏A survey on activity recognition and behavior understanding in video surveillanceVISUAL COMPUTER
VISUAL COMPUTERIF2.9
PREAI
分享收藏Long-Term Effects of Cathodic Protection on Prestressed Concrete Structures: Hydrogen Embrittlement of Prestressing SteelCORROSION
CORROSIONIF0
PREAI
分享收藏
查看更多内容