Shape2Motion: Joint Analysis of Motion Parts and Attributes from 3D Shapes

This is the code repository for "Shape2Motion: Joint Analysis of Motion Parts and Attributes from 3D Shapes”

Created by Xiaogang Wang, Bin Zhou, Yahao Shi, Xiaowu Chen, Qinping Zhao, Kai Xu

Prerequisites:

Numpy (ver. 1.13.3)
TensorFlow (ver. 1.4.1)
scipy (ver. 0.19.1)

Dataset

[June 13, 2019] We release our 3D object mobility dataset 'Motion Dataset V0'.

Download link for outside China here.

Download link for China mainland here. password: wyqf

You can download the data (point cloud data) here for training and testing. password: ihlv

Train：

to train Motion Part Proposal Module and Motion Attribute Proposal Module

python train_stage_12 --stage=1

to train a Proposal Matching Module

python train_stage_12 --stage=2 --batch_size=32

to train a Motion Optimizatoin Network

python train_stage_3

Test：

to test Motion Part Proposal Module and Motion Attribute Proposal Module

python test_stage_12 --stage=1

to test a Proposal Matching Module

python test_stage_12 --stage=2 --batch_size=32

to test a Motion Optimizatoin Network

python test_stage_3

Other:

nms.m: This file is used for merging proposal.
generate_stage_2_train_data.m: This file is used for generating stage 2 training_data.
generate_stage_3_train_data.m: This file is used for generating stage 3 training_data.

Evaluation:

evaluation.m: This file is used for part mobility metrics.

Citation

If you find our paper useful in your research, please cite:

@article{wang_siga18,
title = {Shape2Motion: Joint Analysis of Motion Parts and Attributes from 3D Shapes},
author = {Xiaogang Wang and Bin Zhou and Yahao Shi and Xiaowu Chen and Qinping Zhao and Kai Xu},
journal = {IEEE Conference on Computer Vision and Pattern},
volume = {XX},
number = {XX},
pages = {to appear},
year = {2019}
}