软件导刊2023，Vol.22Issue(12)：78-83,6.DOI:10.11907/rjdk.222474
基于注意力机制上下文建模的弱监督动作定位
Weakly Supervised Action Localization Based on Attention Mechanism Context Modeling
摘要
Abstract
Weakly supervised action localization detect the temporal boundaries of action instances and identify their corresponding action cat-egories with only video-level labels.Due to the lack of frame-level classification labels,weakly-supervised action localization has the prob-lems that some action frames with inconspicuous features are difficult to identify and action frames as well as context frames in videos are easily confused.To address both problems,a weakly supervised action localization method based on attention mechanism context modeling is pro-posed.This method added semi-soft attention on the basis of action-backgroud attention for guiding the model to focus on frames with insignifi-cant action features;To separate action frames and context frames,our method utilized context attention for modeling video contextual infor-mation.The experimental results show that our proposed method has better action localization effect.When the IoU(Intersection over Union)value is 0.5,the average detection accuracy(mAP)on the THUMOS14 and ActivityNet1.3 public datasets reach 32.6%and 38.6%respective-ly,which is better than existing weakly supervised action localization models.关键词
弱监督/动作定位/注意力机制/半软注意力/上下文建模Key words
weakly supervised/action localization/attention mechanism/semi-soft attention/context modeling分类
信息技术与安全科学引用本文复制引用
党伟超,王飞,高改梅,刘春霞..基于注意力机制上下文建模的弱监督动作定位[J].软件导刊,2023,22(12):78-83,6.基金项目
太原科技大学博士科研启动基金项目(20202063) （20202063）
太原科技大学研究生教育创新项目(SY2022063) （SY2022063）
