0
点赞
收藏
分享

微信扫一扫

论文阅读 (37):Discriminative multi-instance multitask learning for 3D action recognition

西特张 2022-01-16 阅读 27

文章目录

引入

  题目用于3D动作识别的辨别性多示例多任务学习 (Discriminative multi-instance multitask Learning for 3D action recognition)
  代码
  摘要:随着低成本和易操作深度相机的蓬勃发展,基于骨骼的人体动作识别得到了广泛研究。然而,大多数现有方法简单地认为人体骨骼的所有3D关节都是相同的。事实上,这些3D关节对不同的动作类型会表现出不同的响应,并且一些关节配置更能区分某个动作。本文提出辨别性多示例多任务学习 (discriminative multi-instance multitask learning, MIMTL)框架,用以发掘关节配置和动作类型之间的本质关系:
  1)将动作和关节配置分别视作多示例学习类型的包及实例,用于捕获一组用于相应动作类型的辨别性和新信息关节配置;
  2)利用具有组结构约束的多任务学习模型来揭示关节配置与不同动作类别之间的内在联系。
  Bib

@article{Yang:2017:519529,  
author		=	{Yan Hua Yang and Cheng Deng and Shang Qian Gao and Wei Liu and Da Peng Tao and Xin Bo Gao},
title		=	{Discriminative multi-instance multitask learning for $3${D} action recognition},
journal		=	{{IEEE} Transactions on Multimedia},
volume		=	{19},
number		=	{3},
year		=	{2017},
pages		=	{519-529}
}

  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  

举报

相关推荐

0 条评论