Skip to main navigation Skip to search Skip to main content

Dynamic Graph CNN with Attention Module for 3D Hand Pose Estimation

  • Xu Jiang
  • , Xiaohong Ma*
  • *Corresponding author for this work
  • Dalian University of Technology

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Recently, 3D hand pose estimation methods taking point cloud as input show the most advanced performance. We present a new 3D deep learning hand pose estimation network for an unordered point cloud. Our approach utilizes EdgeConv layer as the basic element, where an attention embedding version EdgeConv layer is proposed for feature extraction in hand pose estimation task. To improve the result, we design a hand pose improvement network that inputs points whose are in the neighbor of the estimated fingers and outputs a rectify hand pose. We evaluate our method on several famous datasets to prove that our method can get excellent result compared to some most advanced methods.

Original languageEnglish
Title of host publicationAdvances in Neural Networks – ISNN 2019 - 16th International Symposium on Neural Networks, ISNN 2019, Proceedings
EditorsHuchuan Lu, Huajin Tang, Zhanshan Wang
PublisherSpringer Verlag
Pages87-96
Number of pages10
ISBN (Print)9783030227951
DOIs
StatePublished - 2019
Externally publishedYes
Event16th International Symposium on Neural Networks, ISNN 2019 - Moscow, Russian Federation
Duration: 10 Jul 201912 Jul 2019

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11554 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference16th International Symposium on Neural Networks, ISNN 2019
Country/TerritoryRussian Federation
CityMoscow
Period10/07/1912/07/19

Keywords

  • 3D hand pose estimation
  • Attention embedding module
  • Point cloud

Fingerprint

Dive into the research topics of 'Dynamic Graph CNN with Attention Module for 3D Hand Pose Estimation'. Together they form a unique fingerprint.

Cite this