人工智能培训

搜索

深度学习论文:POINTNETLK:使用POINTNET进行强大而高效的点云注册(PointNetLK: Robust & Efficient Point Cl

[复制链接]
einter 发表于 2019-3-15 12:04:31 | 显示全部楼层 |阅读模式
einter 2019-3-15 12:04:31 439 0 显示全部楼层
深度学习论文:POINTNETLK:使用POINTNET进行强大而高效的点云注册(PointNetLK: Robust & Efficient Point Cloud Registration using PointNet)PointNet彻底改变了我们对点云表示的看法。对于分类和分段任务,该方法及其子序列扩展是最先进的。迄今为止,PointNet成功应用于点云注册仍然难以实现。在本文中,认为PointNet本身可以被认为是一种可学习的“成像”功能。因此,用于图像对齐的经典视觉算法可以应用于该问题 - 即Lucas& amp; Kanade(LK)算法。我们的中心创新源于:(i)如何修改LK算法以适应PointNet成像功能,以及(ii)将PointNet和LK算法展开到单个可训练的复发深度神经网络。我们描述了架构,并将其性能与通用注册场景中的最新技术进行了比较。该体系结构提供了一些显着的属性,包括:形状类别的泛化和计算效率 - 为深度学习的topoint云注册应用开辟了新的探索路径。代码和视频可通过此https网址获得。
PointNet has revolutionized how we think about representing point clouds.Forclassification and segmentation tasks, the approach and its subsequentextensions are state-of-the-art.To date, the successful application ofPointNet to point cloud registration has remained elusive.In this paper weargue that PointNet itself can be thought of as a learnable "imaging" function.As a consequence, classical vision algorithms for image alignment can beapplied on the problem - namely the Lucas &Kanade (LK) algorithm.Our centralinnovations stem from: (i) how to modify the LK algorithm to accommodate thePointNet imaging function, and (ii) unrolling PointNet and the LK algorithminto a single trainable recurrent deep neural network.We describe thearchitecture, and compare its performance against state-of-the-art in commonregistration scenarios.The architecture offers some remarkable propertiesincluding: generalization across shape categories and computational efficiency- opening up new paths of exploration for the application of deep learning topoint cloud registration.Code and videos are available atthis https URL.深度学习论文:POINTNETLK:使用POINTNET进行强大而高效的点云注册(PointNetLK: Robust & Efficient Point Cloud Registration using PointNet) ay7u7FUUK6Yyl5Lc.jpg
URL地址:https://arxiv.org/abs/1903.05711     ----pdf下载地址:https://arxiv.org/pdf/1903.05711    ----深度学习论文:POINTNETLK:使用POINTNET进行强大而高效的点云注册(PointNetLK: Robust & Efficient Point Cloud Registration using PointNet)
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

einter当前离线
新手上路

查看:439 | 回复:0

快速回复 返回顶部 返回列表