3D Deep Learning papers review

papers

Name	Paper	Abstract	code
2017
PointNet	[CVPR] PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation.	多层感知, point-wise, maxpool实现对称函数（置换不变）detail seg. cla. 3Dconv	tensorflow
PointNet++	[NeurlIPS] PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space	local+global; Multi-scale; Multi-resolution; detail seg. cla. 3Dconv	tensorflow pytorch
2018
SPG	[CVPR] Large-scale Point Cloud Semantic Segmentation with Superpoint Graphs	graph-based 非监督的超点 Gate graph NN ECC，以Pointnet为基本计算单元	pytorch
2019
KPConv	[ICCV] KPConv: Flexible and Deformable Convolution for Point Clouds	基于位置的卷积 学习卷积核点位置核权值 detail seg. cla. 3Dconv	tensorflow
SPH3D	1. Octree guided CNN with Spherical Kernels for 3D Point Clouds 2. Spherical Kernel for EfficientGraph Convolution on 3D Point Clouds	3DSC bin partition 1. Octree-based 2. Graph-based 球卷积 seg. cls.	code
DeformFilters	Deformable Filter Convolution for Point Cloud Reasoning	类似于KPConv,需详细阅读参考文献[9,16,19,31,38,42]
LPD-Net	[ICCV] LPD-Net: 3D Point Cloud Learning for Large-Scale Place Recognition and Environment Analysis	hand-crafted feature 特征相关 图构建MLP place recognition large-scale  detail