admin管理员组

文章数量:1650765

文章目录

  • PV-RCNN
    • RPN
      • Backbone: 3D Sparse Convolution
      • Classification & Regression Head
    • Voxel Set Abastraction Module(VSA)
      • Discussion
      • VSA Module
      • PKW Module(Predicted Keypoint Weighting)
    • RCNN
  • Experiments
  • 思考

本论文目前是KITTI排名第一,香港中文大学和商汤出品,该作者还提出了PointRCNN和Part-A 2 ^2 2Net。

PV-RCNN

本文将Grid-based(我一般常称为Voxel-based)的方法和Point-based的方法优缺点结合了起来。本文首先说明了Grid-based和Point-based的方法的优缺点:
“Generally, the grid-based methods are more computationally efficient but the inevitable information loss degrades the fine- grained localization accuracy, while the point-based methods have higher computation cost but could easily achieve larger receptive field by the point set abstraction.”

网络的结构图如下:

RPN

Backbone: 3D Sparse Convolution

在本文中没有介绍太多,但在作者之前的一篇文章“Part-A 2 ^2 2 Net: 3D Part-Aware and Aggregation Neural Network for Object”中介绍的比较详细,由于是ba

本文标签: 目标论文RCNNPointVoxel