Skip to content
/ SPoTr Public

Official pytorch implementation of "Self-positioning Point-based Transformer for Point Cloud Understanding" (CVPR 2023).

Notifications You must be signed in to change notification settings

mlvlab/SPoTr

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Self-positioning Point-based Transformer for Point Cloud Understanding

Official pytorch implementation of "Self-positioning Point-based Transformer for Point Cloud Understanding" (CVPR 2023).

Figure

Jinyoung Park 1* , Sanghyeok Lee 1* , Sihyeon Kim 1 , Yunyang Xiong 2 , Hyunwoo J. Kim 1†
1 Korea University, 2 Meta Reality Labs

Setup

  • Clone repository
git clone https://github.com/mlvlab/SPoTr.git
cd SPoTr
  • Install packages with a setup file
bash install.sh
  • Dataset
mkdir -p data/S3DIS/
cd data/S3DIS
gdown https://drive.google.com/uc?id=1MX3ZCnwqyRztG1vFRiHkKTz68ZJeHS4Y
tar -xvf s3disfull.tar
cd ../../

cd data && mkdir ShapeNetPart && cd ShapeNetPart
gdown https://drive.google.com/uc?id=1W3SEE-dY1sxvlECcOwWSDYemwHEUbJIS
tar -xvf shapenetcore_partanno_segmentation_benchmark_v0_normal.tar
cd ../../

Run Experiments

If you want pretrained models, you can download pretrained models via this URL

SNPart

  • Train
CUDA_VISIBLE_DEVICES='0' python examples/shapenetpart/main.py --cfg cfgs/shapenetpart/spotr.yaml
  • Inference
CUDA_VISIBLE_DEVICES='0' python examples/shapenetpart/main.py --cfg cfgs/shapenetpart/spotr.yaml mode=test --pretrained_path ckpt/ShapeNetPart/ckpt_best.pth

S3DIS

  • Train
CUDA_VISIBLE_DEVICES='0' python examples/segmentation/main.py  --cfg cfgs/s3dis/spotr.yaml
  • Inference
CUDA_VISIBLE_DEVICES='0' python examples/segmentation/main.py  --cfg cfgs/s3dis/spotr.yaml  mode=test --pretrained_path ckpt/S3DIS/ckpt_best.pth

Acknowledgement

This repo is built upon OpenPoints.

https://github.com/guochengqian/openpoints

About

Official pytorch implementation of "Self-positioning Point-based Transformer for Point Cloud Understanding" (CVPR 2023).

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published