POS-BERT: Point Cloud One-Satge BERT Pre-training

PyTorch implementation and pretrained models for POS-BERT.

Install

Please install PyTorch. This codebase has been developed with python version 3.8.10, PyTorch version 1.8.0, CUDA 10.2 and torchvision 0.9.0.

conda create -n posbert python=3.8
conda activate posbert
# follow PyTorch website install torch=1.8.0
pip -r requirements.txt
cd install/pointnet2_ops_lib
python setup.py install

Model weights

download the model weights from Google Drive and put it in the weight folder.

Pretrain Eval

experiment results link to Table1 (ET1)

chmod a+x 1exp_pretrain_eval_svm.sh
./1exp_pretrain_eval_svm.sh
./run.sh

Downstream Tasks

cd ../segmentation

License

This repository is released under the Apache 2.0 license as found in the LICENSE file.

Citation

If you find this repository useful, please consider giving a citation:

@article{eswa2023posbert,
  title={POS-BERT: Point Cloud One-Stage BERT Pre-Training},
  author={Kexue Fu, Peng Gao, Shaolei Liu, Linhao Qu, Longxiang Gao, Manning Wang},
  journal={Expert Systems With Applications},
  year={2023}
}