人工智能培训

搜索

深度学习论文:基于残余置换等变层的点姿态投票手势估计(Point-to-Pose Voting based Hand Pose Estimation using

[复制链接]
firetea 发表于 2018-12-6 12:00:04 | 显示全部楼层 |阅读模式
firetea 2018-12-6 12:00:04 128 0 显示全部楼层
深度学习论文:基于残余置换等变层的点姿态投票手势估计(Point-to-Pose Voting based Hand Pose Estimation using Residual  Permutation Equivariant Layer)最近,基于3D输入数据的手姿势估计方法已经显示出最先进的性能,因为3D数据捕获了比深度图像更多的空间信息。虽然基于3D体素的方法需要大量内存,但基于PointNet的方法需要繁琐的预处理步骤,例如每个点的K最近邻搜索。在本文中,我们提出了一种新的深度学习手姿态估计方法,用于无序点云。 Ourmethod需要1024个3D点作为输入,不需要额外的信息。我们使用置换等变层(PEL)作为基本元素,其中为手姿态估计任务提出了PEL的剩余网络版本。此外,我们提出了一种基于投票的方案,用于将来自各个点的信息合并到最终姿势输出。除了参数估计任务之外,基于投票的方案还可以提供点云分割结果而没有用于分割的基础事实。我们在NYU数据集和Hands2017Challenge数据集上评估我们的方法。我们的方法优于最新的最先进方法,其中我们的姿势精度目前是Hands2017Challenge数据集的最佳选择。
Recently, 3D input data based hand pose estimation methods have shownstate-of-the-art performance, because 3D data capture more spatial informationthan the depth image.Whereas 3D voxel-based methods need a large amount ofmemory, PointNet based methods need tedious preprocessing steps such asK-nearest neighbour search for each point.In this paper, we present a noveldeep learning hand pose estimation method for an unordered point cloud.Ourmethod takes 1024 3D points as input and does not require additionalinformation.We use Permutation Equivariant Layer (PEL) as the basic element,where a residual network version of PEL is proposed for the hand poseestimation task.Furthermore, we propose a voting based scheme to mergeinformation from individual points to the final pose output.In addition to thepose estimation task, the voting-based scheme can also provide point cloudsegmentation result without ground-truth for segmentation.We evaluate ourmethod on both NYU dataset and the Hands2017Challenge dataset.Our methodoutperforms recent state-of-the-art methods, where our pose accuracy iscurrently the best for the Hands2017Challenge dataset.深度学习论文:基于残余置换等变层的点姿态投票手势估计(Point-to-Pose Voting based Hand Pose Estimation using Residual  Permutation Equivariant Layer) fyuUUU38G243qyZ8.jpg
URL地址:https://arxiv.org/abs/1812.02050     ----pdf下载地址:https://arxiv.org/pdf/1812.02050    ----深度学习论文:基于残余置换等变层的点姿态投票手势估计(Point-to-Pose Voting based Hand Pose Estimation using Residual  Permutation Equivariant Layer)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

firetea当前离线
新手上路

查看:128 | 回复:0

快速回复 返回顶部 返回列表