Point Cloud Architectures - implementations

Implementation of SOTA Point Cloud Deep Learning Networks Using TensorFlow(TF) 2 or Pytorch

Inspired by Machine-Learning-Tokyo/CNN-Architectures, this repo is to show how to implement SOTA point cloud deep learning-based networks with the tensorflow.keras Functional API based on the original paper. For each architecture there is:

• the paper in which the architecture was originally published where we highlighted the parts needed for the implementation
• a jupyter notebook with a step-by-step description of how to use this information to infer the structure of the model and code it
• a diagram of the model that corresponds to the architecture and the code structure

Contents

• PointNet [Notebook]

• PointNet++ [Notebook]

• KPConv [Notebook]

• RandLA-Net [Notebook]

Resources

PointNet family

• charlesq34/pointnet, official implementation in TF 1.
• charlesq34/pointnet2, official implementation in TF 1.
• yanx27/Pointnet_Pointnet2_pytorch
• fxia22/pointnet.pytorch
• erikwijmans/Pointnet2_PyTorch

KPConv

• HuguesTHOMAS/KPConv, official implementation in TF 1.
• HuguesTHOMAS/KPConv-PyTorch, official implementation in Pytorch.

RandLA-Net

• QingyongHu/RandLA-Net), official implementation in TF 1.