[骨架动作识别]ST-NBNN&Deep Learning on Lie Groups CVPR2017

本文介绍了ST-NBNN方法,它结合参数和非参数模型的优势,通过stage-to-class距离和双线性分类器进行骨架动作识别。此外,还提出了一种基于李群的神经网络LieNet,用于学习3D骨架特征。LieNet通过调整网络结构以适应李群,实现了对动作识别的改进。实验结果显示,这两种方法在动作识别上表现优秀,甚至仅使用线性分类器就能超越某些端到端模型。

摘要生成于 C知道 ,由 DeepSeek-R1 满血版支持, 前往体验 >

一、ST-NBNN:

没上神经网络
Each 3D action instance is represented by a collection of temporal stages composed by 3D poses, and each pose in stages is presented by a collection of spatial joints.

1. Introduction

learning-based classifiers,基于学习的骨架分类方法[5, 24, 35, 21, 14, 13] 已经取得了很多进展
non-parametric classifiers 不需要学习,训练参数 的分类方法还没有被很好的探索

  • 动机有二
    1:
    2:骨架信息并不像图片,有成千上万个像素,只有上十个joints,因此不需要端到端的复杂模型,非参数模型也能搞定
    这里写图片描述
  • 方法
    1:通过使用stage-to-class distance and bilinear classifier,该模型结合了参数模型和非参数模型的长处
    2:通过关键帧和关键的骨架节点(key temporal stages and spatial joints),模型能提取出必要的时空模型spatio-temporal patterns

  • 结果
    仅用线性分类器就能超过很多端到端的模型

李群的一本书,是扫描版,书的质量不错。 This book is intended for a one year graduate course on Lie groups and Lie algebras. The author proceeds beyond the representation theory of compact Lie groups (which is the basis of many texts)and provides a carefully chosen range of material to give the student the bigger picture. For compact Lie groups, the Peter-Weyl theorem, conjugacy of maximal tori (two proofs), Weyl character formula and more are covered. The book continues with the study of complex analytic groups, then general noncompact Lie groups, including the Coxeter presentation of the Weyl group, the Iwasawa and Bruhat decompositions, Cartan decomposition, symmetric spaces, Cayley transforms, relative root systems, Satake diagrams, extended Dynkin diagrams and a survey of the ways Lie groups may be embedded in one another. The book culminates in a "topics" section giving depth to the student's understanding of representation theory, taking the Frobenius-Schur duality between the representation theory of the symmetric group and the unitary groups as a unifying theme, with many applications in diverse areas such as random matrix theory, minors of Toeplitz matrices, symmetric algebra decompositions, Gelfand pairs, Hecke algebras, representations of finite general linear groups and the cohomology of Grassmannians and flag varieties.   Daniel Bump is Professor of Mathematics at Stanford University. His research is in automorphic forms, representation theory and number theory. He is a co-author of GNU Go, a computer program that plays the game of Go. His previous books include Automorphic Forms and Representations (Cambridge University Press 1997)and Algebraic Geometry (World Scientific 1998).
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值