xyf513 / SpiderCNN

SpiderCNN: Deep Learning on Point Sets with Parameterized Convolutional Filters

SpiderCNN

SpiderCNN: Deep Learning on Point Sets with Parameterized Convolutional Filters. ECCV 2018
Yifan Xu, Tianqi Fan, Mingye Xu, Long Zeng, Yu Qiao.

Introduction

This project is based on our ECCV18 paper. You can find the arXiv version here.

@article{xu2018spidercnn,
  title={SpiderCNN: Deep Learning on Point Sets with Parameterized Convolutional Filters},
  author={Xu, Yifan and Fan, Tianqi and Xu, Mingye and Zeng, Long and Qiao, Yu},
  journal={arXiv preprint arXiv:1803.11527},
  year={2018}
}

SpiderCNN is a convolutional neural network that can process signals on point clouds.

Installation

The code is based on PointNet， and PointNet++. Please install TensorFlow, and follow the instruction in PointNet++ to compile the customized TF operators.
The code has been tested with Python 2.7, TensorFlow 1.3.0, CUDA 8.0 and cuDNN 6.0 on Ubuntu 14.04.

Usage

Classification

Preprocessed ModelNet40 dataset can be downloaded here.
To train a SpiderCNN model (with input XYZ coordinates and normal vectors) to classify shapes in ModelNet40:

python train.py

To train a SpiderCNN model (with input XYZ coordinates) with multi GPU to classify shapes in ModelNet40:

python train_xyz.py

Part Segmentation

Preprocessed ShapeNetPart dataset can be downloaded here. To train a model to segment object parts for ShapeNet models (with input XYZ coordinates and normal vectors):

cd part_seg
python train.py

License

This repository is released under MIT License (see LICENSE file for details).