Chinese Journal of Liquid Crystals and Displays, Volume. 37, Issue 6, 736(2022)
3D hand pose estimation algorithm based on cascaded features and graph convolution
Figures & Tables(9)
Fig. 4. Loss function curves during network training and AUC accuracy curves under the validation set
Table 1. Distribution of data used in the experiment
View tableView in ArticleTable 1. Distribution of data used in the experiment
手部2D标注 手部3D标注 训练集 测试集 合计 - - 389 477 9 063 CMU［ 14 ］√ - 16 720 - RHD［ 15 ］√ √ 41 258 2 728 Dexter+Object［ 16 ］√ √ - 3 145 EgoDexter［ 17 ］- √ - 3 190 GANeratedDataset［ 18 ］- √ 331 499 -
Table 2. Comparison of experimental results between this method and other methods
View tableView in ArticleTable 2. Comparison of experimental results between this method and other methods
方法
数据
Dexter+Object［ 16 ］EgoDexter［ 17 ］RHD［ 15 ］20 mm 30 mm AUC 20 mm 30 mm AUC 20 mm 30 mm AUC Mesh2HAND［ 19 ］0.732 0.856 0.825 0.717 0.744 0.746 0.800 0.915 0.901 Mon-Cap［ 20 ］0.795 0.933 0.912 0.732 0.811 0.779 0.816 0.923 0.911 Adane［ 21 ］0.653 0.819 0.763 0.636 0.701 0.674 0.798 0.902 0.887 Minimal-hand［ 5 ］0.821 0.953 0.923 0.838 0.917 0.914 0.799 0.883 0.856 本文方法 0.830 0.967 0.931 3 0.821 0.954 0.921 3 0.824 0.961 0.934 9
Table 3. Comparison of the time of inference and FLOPs
View tableView in ArticleTable 3. Comparison of the time of inference and FLOPs
推理时间/ms 计算量（FLOPs） Mesh2Hand［ 19 ］1 173 11.3E10 Mon-Cap［ 20 ］564 7.9E9 Adanet［ 21 ］642 8.5E9 Minimal-hand［ 5 ］49 8E8 本文方法 52 9.3E8
Tools
Get Citation
Copy Citation Text
Yi-lin LIN, Shan-ling LIN, Zhi-xian LIN. 3D hand pose estimation algorithm based on cascaded features and graph convolution[J]. Chinese Journal of Liquid Crystals and Displays, 2022, 37(6): 736
Paper Information
Category:
Received: Nov. 29, 2021
Accepted: --
Published Online: Jun. 22, 2022
The Author Email: Zhi-xian LIN (lzx2005000@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20