Point Cloud Semantic Segmentation by Adaptively Fusing Information with Varying Distances

by Zefeng Jiang, Baochen Yao, Kangkang Song, Xiaojie Qiu, and Chengbin Peng, details are in paper.

In this work, we argue that adaptive varying-distance feature aggregation and discrimination can improve the effect of point cloud semantic segmentation. The proposed approach consists of three steps. First, we agglomerate points into superpoints and construct a superpoint graph as many traditional approaches. Second, we propose a novel varying-distance autoencoder to help each superpoint adaptively assimilate information from different distances. Third, we propose a discrimination loss to constrain the embedding space so that superpoints belonging to the same semantic class can get closer and vice versa.

Dataset:

Download S3DIS Dataset and extract Stanford3dDataset_v1.2_Aligned_Version.zip to $S3DIS_DIR/data, where $S3DIS_DIR is set to dataset directory.

Requirements:

• environment:

  Ubuntu 20.04
  

• install python package:

  ./Anaconda3-5.1.0-Linux-x86_64.sh
  

• install PyTorch :

  conda install pytorch==1.11.0
  

Train

python learning/main.py

Test

python learning/main.py --epochs -1 --resume RESUME

Citation:

 @article{jiang2024point,
  title={Point Cloud Semantic Segmentation by Adaptively Fusing Information With Varying Distances},
  author={Jiang, Zefeng and Yao, Baochen and Song, Kangkang and Qiu, Xiaojie and Peng, Chengbin},
  journal={IEEE Signal Processing Letters},
  year={2024},
  publisher={IEEE}
}