qianyun-li / soe-net

The codes for CVPR2021 Oral paper 'SOE-Net: A Self-Attention and Orientation Encoding Network for Point Cloud based Place Recognition'

SOE-Net: A Self-Attention and Orientation Encoding Network for Point Cloud based Place Recognition (CVPR 2021 Oral)

This repository is the official implementation for paper:

SOE-Net: A Self-Attention and Orientation Encoding Network for Point Cloud based Place Recognition

Yan Xia, Yusheng Xu, Shuang Li, Rui Wang, Juan Du, Daniel Cremers, Uwe Stilla

Technical University of Munich, Beijing Insitute of Technology, Artisense

Introduction

---

SOE-Net fully explores the relationship between points and incorporates long-range context into point-wise local descriptors. Local information of each point from eight orientations is captured in a PointOE module, whereas long-range feature dependencies among local descriptors are captured with a self-attention unit. Moreover, we propose a novel loss function called Hard Positive Hard Negative quadruplet loss (HPHN quadruplet), that achieves better performance than the commonly used lazy quadruplet loss.

Pre-requisites

---

• Python3.6
• Tensorflow1.4.0
• CUDA-9.0
• Scipy
• Pandas
• Sklearn
• The TF operators under tf_ops folder should be compiled.
• generate pickle files, refer to PointNetVLAD.

Training

---

python train.py

Evaluation

---

python evaluate.py

Pretrained models

---

The pre-trained models for both the baseline and refinement networks can be downloaded here.

Acknowledgement

---

The code is in heavily built on PointNetVLAD. We also borrow something from PointSIFT.

Citation

---

If you find our work useful in your research, please consider citing:

@inproceedings{xia2021soe,
 author = {Y. Xia and Y. Xu and S. Li and R. Wang and J. Du and D. Cremers and U. Stilla},
 title = {SOE-Net: A Self-Attention and Orientation Encoding Network for Point Cloud based Place Recognition},
 booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
 year = {2021},
 award = {Oral Presentation},
}