hanmakaidao/sqn

SQN in Tensorflow (ECCV'2022)

SQN: Weakly-Supervised Semantic Segmentation of Large-Scale 3D Point Clouds (ECCV2022)

This is the official repository of the Semantic Query Network (SQN). For technical details, please refer to:

SQN: Weakly-Supervised Semantic Segmentation of Large-Scale 3D Point Clouds
Qingyong Hu, Bo Yang, Guangchi Fang , Ales Leonardis, Yulan Guo, Niki Trigoni , Andrew Markham.
[Paper] [Video]

(1) Setup

This code has been tested with Python 3.5, Tensorflow 1.11, CUDA 9.0 and cuDNN 7.4.1 on Ubuntu 16.04/Ubuntu 18.04.

• Clone the repository

git clone --depth=1 https://github.com/QingyongHu/SQN && cd SQN

• Setup python environment

conda create -n sqn python=3.5
source activate sqn
pip install -r helper_requirements.txt
sh compile_op.sh

(2) Training (Semantic3D as example)

First, follow the RandLA-Net instruction to prepare the dataset, and then manually change the dataset path here.

• Start training with weakly supervised setting:

python main_Semantic3D.py --mode train --gpu 0 --labeled_point 0.1%

• Evaluation:

python main_Semantic3D.py --mode test --gpu 0 --labeled_point 0.1%

Quantitative results achieved by our SQN:

(3) Sparse Annotation Demo

Citation

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

@inproceedings{hu2021sqn,
  title={SQN: Weakly-Supervised Semantic Segmentation of Large-Scale 3D Point Clouds},
  author={Hu, Qingyong and Yang, Bo and Fang, Guangchi and Guo, Yulan and Leonardis, Ales and Trigoni, Niki and Markham, Andrew},
  booktitle={European Conference on Computer Vision},
  year={2022}
}

Related Repos

1. RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds
2. SoTA-Point-Cloud: Deep Learning for 3D Point Clouds: A Survey
3. 3D-BoNet: Learning Object Bounding Boxes for 3D Instance Segmentation on Point Clouds
4. SpinNet: Learning a General Surface Descriptor for 3D Point Cloud Registration
5. SensatUrban: Learning Semantics from Urban-Scale Photogrammetric Point Clouds
6. Not All Points Are Equal: Learning Highly Efficient Point-based Detectors for 3D LiDAR Point Clouds