Code for "Point cloud processing via recurrent set encoding".

Requirements:

• PyTorch 0.3/0.4
• Python 3.6+
• CUDA 8.0 (Not sure if CUDA > 8.0 will work. This depends on PyTorch.)

Usage and file structures:

• For classification, run train_rcnet_cls.py. For shape segmentation, run run_seg.py
• The dataset can be downloaded from https://github.com/charlesq34/pointnet:

# classification data
wget https://shapenet.cs.stanford.edu/media/modelnet40_normal_resampled.zip

# segmentation data
wget https://shapenet.cs.stanford.edu/media/shapenetcore_partanno_segmentation_benchmark_v0_normal.zip

• The dataloader for classification task is in file utils/datasets.py. The dataloader for segmentation task is in file utils/part_dataset.py.
• utils/pointnet.py1 provides some modules, which are about spatial transformer (STN).
• utils/orderpoints.py provides functions for partitioning the ambient space into structured beam. Currently the code is not the most efficient. For more efficient version, please refer to here.
• utils/provider provides functions for some basic data augmentation, such at random jitter, random scale, etc.
• Directory utils/gen_point_cloud includes some codes (you may not need) for converting mesh into point clouds. It relies on pyntcloud library.

Reference:

@inproceedings{wu2019point,
  title={Point cloud processing via recurrent set encoding},
  author={Wu, Pengxiang and Chen, Chao and Yi, Jingru and Metaxas, Dimitris},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={33},
  pages={5441--5449},
  year={2019}
}