Awesome

PointWeb: Enhancing Local Neighborhood Features for Point Cloud Processing

by Hengshuang Zhao*, Li Jiang*, Chi-Wing Fu, and Jiaya Jia, details are in paper.

Introduction

This repository is build for PointWeb in point cloud scene understanding.

Usage

Requirement:
- Hardware: 4 GPUs (better with >=11G GPU memory)
- Software: PyTorch>=1.0.0, Python3, CUDA>=9.0, tensorboardX

Clone the repository and build the ops:

git clone https://github.com/hszhao/PointWeb.git
cd PointWeb
cd lib/pointops && python setup.py install && cd ../../

Train:
- Download related datasets and symlink the paths to them as follows (you can alternatively modify the relevant paths specified in folder config):
```
mkdir -p dataset
ln -s /path_to_s3dis_dataset dataset/s3dis
```
- Specify the gpu used in config and then do training:
```
sh tool/train.sh s3dis pointweb
```
Test:
- Download trained segmentation models and put them under folder specified in config or modify the specified paths.
- For full testing (get listed performance):
```
sh tool/test.sh s3dis pointweb
```
Visualization: tensorboardX incorporated for better visualization.
```
tensorboard --logdir=run1:$EXP1,run2:$EXP2 --port=6789
```
Other:
- Resources: GoogleDrive LINK contains shared models, predictions and part of the related datasets.
- Video predictions: Youtube LINK.

Performance

Description: mIoU/mAcc/aAcc/voxAcc stands for mean IoU, mean accuracy of each class, all pixel accuracy , and voxel label accuracy respectively.

mIoU/mAcc/aAcc of PointWeb on S3DIS dataset: 0.6055/0.6682/0.8658.

mIoU/mAcc/aAcc/voxAcc of PointWeb on ScanNet dataset: 0.5063/0.6061/0.8529/0.8568.

Citation

If you find the code or trained models useful, please consider citing:

@inproceedings{zhao2019pointweb,
  title={{PointWeb}: Enhancing Local Neighborhood Features for Point Cloud Processing},
  author={Zhao, Hengshuang and Jiang, Li and Fu, Chi-Wing and Jia, Jiaya},
  booktitle={CVPR},
  year={2019}
}