PointRNN: Point Recurrent Neural Network for Moving Point Cloud Processing

PointRNN

Structure

PointGRU

PointLSTM

Moving Point Cloud Prediction

Installation

The code is tested with Red Hat Enterprise Linux Workstation release 7.7 (Maipo), g++ (GCC) 5.3.1, TensorFlow v1.12, CUDA 9.0 and cuDNN v7.4.

Install TensorFlow v1.12:

pip install tensorflow-gpu==1.12

Compile the CUDA layers for PointNet++, which we used for furthest point sampling (FPS) and radius neighbouring search, and Chamfer Distance (CD) and Earth Mover's Distance (EMD):

cd modules/tf_ops/3d_interpolation && make
cd modules/tf_ops/approxmatch && make
cd modules/tf_ops/grouping && make
cd modules/tf_ops/nn_distance && make
cd modules/tf_ops/sampling && make

Before compiling, plese correctly set the CUDA_HOME and CUDNN_HOME in each Makefile under the 3d_interpolation, approxmatch, grouping, nn_distance and sampling directories, resplectively.

CUDA_HOME := /usr/local/cuda-9.0
CUDNN_HOME := /usr/local/cudnn7.4-9.0

Datasets

We provide the test sets for evaluating moving point cloud prediction:

1. Moving MNIST Point Cloud (1 digit)   2. Moving MNIST Point Cloud (2 digits)   3. Argoverse   4. nuScenes

License

The code is released under MIT License.

Citation

If you find our work useful in your research, please consider citing:

@article{fan19pointrnn,
  author    = {Hehe Fan and Yi Yang},
  title     = {PointRNN: Point Recurrent Neural Network for Moving Point Cloud Processing},
  journal   = {arXiv},
  volume    = {1910.08287},
  year      = {2019}
}

Related Repos

1. PointRNN PyTorch implementation: https://github.com/hehefan/PointRNN-PyTorch
2. PointNet++ TensorFlow implementation: https://github.com/charlesq34/pointnet2

Visualization

Dataset	1 MNIST	2 MNIST	Argoverse	Argoverse	nuScenes	nuScenes
Input
Ground truth
PointRNN
PointGRU
PointLSTM