Awesome

Discrete Rotation Equivariance for Point Cloud Recognition

Discrete Rotation Equivariance for Point Cloud Recognition. ICRA 2019, Montreal, Canada. Jiaxin Li, Yingcai Bi, Gim Hee Lee, National University of Singapore

Introduction

This paper and repository propose a rotation equivariant method to improve deep network's performance when the input point cloud is rotated in 2D or 3D.

Installation

Requirements:

Optional dependency:

Faiss GPU support

Dataset

ModelNet40 and MNIST Our dataset is provided by SO-Net:

Google Drive: MNIST, ModelNet40/10

Usage

Compile Cuda Operations

run python3 setup.py install in models/index_mat_ext to install the customized maxpooling layer.

Configurations

To run these tasks, you may need to set the dataset type and path in options.py, by changing the default value of --dataset, --dataroot.

In training, you may want to configure the input rotation settings in options.py

Input rotation augmentation mode: rot_equivariant_mode, supports '2d' / '3d''
Number of descrete rotation: rot_equivariant_no. Supports {1, 4, 6, 9, 12, 24} when rot_equivariant_mode=2d, {1, 4, 12} when rot_equivariant_mode=3d.

Visualization

We use visdom for visualization. Various loss values and the reconstructed point clouds (in auto-encoder) are plotted in real-time. Please start the visdom server before training, otherwise there will be warnings/errors, though the warnings/errors won't affect the training process.

python3 -m visdom.server

The visualization results can be viewed in browser with the address of:

http://localhost:8097

License

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