JSENet: Joint Semantic Segmentation and Edge Detection Network for 3D Point Clouds

Introduction

Implementation of ECCV2020 paper - JSENet: Joint Semantic Segmentation and Edge Detection Network for 3D Point Clouds (arXiv). If you find our work useful in your research, please consider citing:

@inproceedings{hu2020jsenet,
  title={JSENet: Joint Semantic Segmentation and Edge Detection Network for 3D Point Clouds},
  author={Hu, Zeyu and Zhen, Mingmin and Bai, Xuyang and Fu, Hongbo and Tai, Chiew-lan},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  pages={222--239},
  year={2020}
}

Installation

This repository is modified from KPConv, please find the step-by-step installation guide in INSTALL.md.

Experiments

Data

S3DIS

S3DIS dataset can be downloaded here (4.8 GB). Download the file named Stanford3dDataset_v1.2.zip, uncompress the folder and move it to Data/S3DIS.

We provide processed demo dataset for experiments on S3DIS fold-5. The demo dataset can be downloaded here (903 MB). Uncompress the folder and move it to Data/S3DIS.

Scannet

Scannet dataset can be find here. Follow the instructions and move downloaded files to Data/Scannet.

Training

For S3DIS dataset:

python training_S3DIS.py

For Scannet dataset:

python training_Scannet.py

If you are not using the processed demo dataset, the first run will take some time to process the raw data. The process can be easily accelerated using parallel computing methods like Pthreads.

Testing

In test_model.py, you will find detailed comments explaining how to choose which logged trained model you want to test. Follow them and then run the script :

For semantic segmentation task:

python test_model.py --task SS

For semantic edge detection task:

python test_model.py --task SED

Pretrained models

Pretrained models can be downloaded here (119 MB). Uncompress the files and move it to results/.

Acknowledgment

Our code is modified from KPConv.

License

Our code is released under MIT License (see LICENSE file for details).