[中文版]

This repository contains the PyTorch implementation for paper SegGroup: Seg-Level Supervision for 3D Instance and Semantic Segmentation

Authors: An Tao, Yueqi Duan, Yi Wei, Jiwen Lu, Jie Zhou

[arxiv] [Zhihu] [FAQ]

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

@article{tao2022seggroup,
  title={{SegGroup}: Seg-Level Supervision for {3D} Instance and Semantic Segmentation},
  author={Tao, An and Duan, Yueqi and Wei, Yi and Lu, Jiwen and Zhou, Jie},
  journal={IEEE Transactions on Image Processing},
  year={2022},
  volume={31},
  pages={4952-4965},
  publisher={IEEE}
}

Our annotation tool is in antao97/SegGroup.annotator.

Updates:

• [2022/07/01] This work is accepted by IEEE Transactions on Image Processing!
• [2022/06/23] We update our paper in arXiv.

Our seg-level supervised point cloud segmentation method can be divided into two steps: 1) pseudo label generation with SegGroup and 2) fully-supervised point cloud segmentation model training with pseudo labels. The two stages are trained separately, and the evaluation of the segmentation performance is conducted on the model trained in step 2.

Use our designed SegGroup model in seggroup/ to generate point-level pseudo labels from seg-level labels.

After generating pseudo labels, we can use them to replace the ground-truth labels on the training set to train a standard point cloud segmentation model with full supervision.

In our work, our pseudo labels can be used in both instance segmentation and semantic segmentation task.

• pointgroup/

• kpconv/
• minkowski/

• Difference between SegGroup and OTOC?
• Details on over-segmentation?
• How to determine the annotation rule?