在第一阶段作者引入了基于点的球形anchor生成了较为准确的propoals。然后使用PointsPool生成了紧凑的表示,这有利于加速推理。第二阶段减少了错误的删除(应该是指NMS用了更合理的参考IoU)。最后,该模型在3D目标检测任务中表现良好。 笔者: 首先不得不说,不开源还挺难受的。作者希望使用PointNet++强大的表示能力,但是又...
STD:Sparse-to-Dense 3D Object Detector for Point Cloud(腾讯&香港大学),主要思想本文提出了一种新的两阶段3D目标检测框架,称之为稀疏到稠密三维目标检测框架(STD)。第一个阶段是自下而上的proposal生成网络,该网络使用原始点云作为输入,通过为每个点播种新的球形a
【文章阅读】STD:Sparse-to-Dense 3D Object Detector for Point Cloud,程序员大本营,技术文章内容聚合第一站。
We present a new two-stage 3D object detection framework, named sparse-to-dense 3D Object Detector (STD). The first stage is a bottom-up proposal generation network that uses raw point cloud as input to generate accurate proposals by seeding each point with a new spherical anchor. It achieve...
STD: Sparse-to-Dense 3D Object Detector for Point Cloud 阅读笔记,程序员大本营,技术文章内容聚合第一站。
Given an image pair (IA,IB)(IA,IB) to be matched, we first applied an off-the-shelf keypoint detector to obtain keypoints written as KA={fnA|n=1,…,NA}KA={fAn|n=1,…,NA} for image IAIA. S2DNet regards feature matching as a sparse-to-dense hypercolumn matching problem[19]. ...
An efficient and a powerful object detector is being proposed which can be trained on a single GPU. State-of-the-art algorithms and methodologies such as FPN [5], PAN [6], YOLO [7], GIoU_Loss [8], etc. have been modified and implemented into the design of the proposed object detecto...