PointNAT: Large Scale Point Cloud Semantic Segmentation via Neighbor Aggregation with Transformer

Here is the PyTorch implementation of the paper PointNAT: Large Scale Point Cloud Semantic Segmentation via Neighbor Aggregation with Transformer.

Setup

• 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 ../../

Train

CUDA_VISIBLE_DEVICES=0 bash script/main_segmentation.sh cfgs/s3dis/pointnat.yaml wandb.use_wandb=True

Test

CUDA_VISIBLE_DEVICES='0' bash script/main_segmentation.sh cfgs/s3dis/pointnat.yaml wandb.use_wandb=False mode=test --pretrained_path path/to/pretrained/model.pth

Acknowledgement

This repo is built upon OpenPoints.

https://github.com/guochengqian/openpoints