Skip to content

point_cloud_semantic_segmentation

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

point_cloud_semantic_segmentation

README.md

SMPointSeg

Network Architecture

Setup

1. Clone this repository

git clone https://github.com/LongguangWang/SparseMask && cd SparseMask/point_cloud_semantic_segmentation

2. Install the following pacages

  • cuda == 11.1
  • PyTorch == 1.8.0
  • prefetch_generator == 1.0.1
  • numpy == 1.19.5
  • open3d == 0.12.0
  • scikit-learn == 0.24.2
  • pandas == 1.2.4

Our code is tested with the above environments.

3. Install the required operations

Run sh compile_op.sh to install required opeartions.

S3DIS

1. Prepare the dataset

1.1 Download the S3DIS datset (Stanford3dDataset_v1.2_Aligned_Version.zip) and uncompress it to your_dir_S3DIS.

1.2 Run sh data_prepare_S3DIS.sh to prepare training data. Generated data will be stored in your_dir_S3DIS_original and your_dir_S3DIS_sub0.040. Please update data_dir in the bash file as your_dir_S3DIS.

2. Begin to train

Run sh train_S3DIS.sh. Please update data_dir in the bash file as your_dir_S3DIS_sub0.040.

3. Begin to test

Run sh inference_S3DIS.sh. Please update data_dir in the bash file as your_dir_S3DIS_sub0.040.

4. 6-fold cross validation

Run sh 6_fold_S3DIS.sh. Please update data_dir in the bash file as your_dir_S3DIS_original.

  • Quantitative Results

  • Visual Results

SemanticKITTI

1. Prepare the dataset

1.1 Download the SemanticKITTI dataset (files related to semantic segmentation) and uncompress it to your_dir_SemanticKITTI.

1.2 Run sh data_prepare_SematicKITTI.sh to prepare training data. Generated data will be stored in your_dir_SemanticKITTI_sequences_0.06. Please update data_dir in the bash file as your_dir_SemanticKITTI.

2. Begin to train

Run sh train_SematicKITTI.sh. Please update data_dir in the bash file as your_dir_SemanticKITTI_sequences_0.06.

3. Begin to test

Run sh inference_SematicKITTI.sh. Please update data_dir in the bash file as your_dir_SemanticKITTI_sequences_0.06.

  • Quantitative Results

  • Visual Results

Citation

@Article{Wang2022Exploring,
  author  = {Longguang Wang and Yulan Guo and Xiaoyu Dong and Yingqian Wang and Xinyi Ying and Zaiping Lin and Wei An},
  title   = {Exploring Fine-Grained Sparsity in Convolutional Neural Networks for Efficient Inference},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year    = {2022},
}

Acknowledgements

Part of the code is borrowed from RandLA-Net, RandLA-Net (PyTorch) and KPConv. We thank the authors for sharing the codes.