RPM-Net: Recurrent Prediction of Motion and Parts from Point Cloud
We introduce RPM-Net, a deep learning-based approach which simultaneously infers movable parts and hallucinates their motions from a single, un-segmented, and possibly partial, 3D point cloud shape. RPM-Net is a novel Recurrent Neural Network (RNN), composed of an encoder-decoder pair with interleaved Long Short-Term Memory (LSTM) components, which together predict a temporal sequence ofpointwise displacements for the input point cloud. At the same time, the displacements allow the network to learn movable parts, resulting in a motion-based shape segmentation.
For more details, please refer to our paper.
All codes are tested under Tensorflow 1.13.1 GPU version and Python 3.6 on Ubuntu 14.04.
First, please download data.zip from our project page, and extract to RPM-Net/data/ folder.
Run train.py to train RPM-Net, the model will be saved in RPM-Net/output/ folder.
Before testing, please modify the model_path and eval_dir in test.py, the predicted pointcloud and segmentation files will be saved in RPM-Net/output/YOUR_MODEL_PATH/eval/ folder.
The original dataset is also released here. See readme file inside for details.
Our code is released under MIT License. See LICENSE file for details.
Please cite the paper in your publications if it helps your research:
@article{RPMNet19,
title = {RPM-Net: Recurrent Prediction of Motion and Parts from Point Cloud},
author = {Zihao Yan and Ruizhen Hu and Xingguang Yan and Luanmin Chen and Oliver van Kaick and Hao Zhang and Hui Huang},
journal = {ACM Transactions on Graphics (Proceedings of SIGGRAPH ASIA 2019)},
volume = {38},
number = {6},
pages = {240:1--240:15},
year = {2019},
}