Skip to content
master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

USIP

USIP: Unsupervised Stable Interest Point Detection from 3D Point Clouds. ICCV 2019, Seoul, Korea
Jiaxin Li, Gim Hee Lee
Department of Computer Science, National University of Singapore

@article{li2019usip,
  title={USIP: Unsupervised Stable Interest Point Detection from 3D Point Clouds},
  author={Li, Jiaxin and Lee, Gim Hee},
  journal={arXiv preprint arXiv:1904.00229},
  year={2019}(
}

Introduction

USIP detector: an Unsupervised Stable Interest Point detector that can detect highly repeatable and accurately localized keypoints from 3D point clouds under arbitrary transformations without the need for any ground truth training data.

Examples of keypoints detected by our USIP are shown below for ModelNet40, Redwood, Oxford RobotCar, KITTI, respectively.

ModelNet40 Redwood Oxford RobotCar KITTI

This repository consists of point cloud keypoint detector / descriptor for Oxford RobotCar, KITTI, SceneNN, ModelNet40, 3DMatch.

Installation

Rquirements:

cd models/index_max_ext
python3 setup.py install
  • Compile customized cuda module - ball_query:
cd models/ball_query_ext
python3 setup.py install
python3 setup.py install

Dataset

This Google Drive link contains datasets used in our paper: KITTI, Oxford RobotCar, 3DMatch, ModelNet40, Redwood, SceneNN.

Trained Models

This Google Drive link contains our trained models for some datasets.

Usage

Each folder of kitti, match3d, modelnet, oxford, scenenn contains configuration script ooptions_*** and training scripts train_***. Please modify the configurations files before running the training scripts. For example, you may have to modify the default value of --dataset, --dataroot, --gpu_ids.

Evaluation

  1. Save detected keypoints via evaluation/save_keypoints.py
  2. Run Matlab based evaluation code via evaluation/matlab

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 GPL-3.0 License (see LICENSE file for details).

About

USIP: Unsupervised Stable Interest Point Detection from 3D Point Clouds

Resources

License

Releases

No releases published

Packages

No packages published