Repository files navigation

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

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

1. Requirement:

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

2. Clone the repository and build the ops:

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

3. Train:

   • Download relateddatasets and symlink the paths to them as follows (you can alternatively modify the relevant paths specified in folderconfig):

     mkdir -p datasetln -s /path_to_s3dis_dataset dataset/s3dis

   • Specify the gpu used in config and then do training:

     sh tool/train.sh s3dis pointweb

4. 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

5. Visualization:tensorboardX incorporated for better visualization.

   tensorboard --logdir=run1:$EXP1,run2:$EXP2 --port=6789

6. Other:

   • Resources: GoogleDriveLINK contains shared models, predictions and part of the related datasets.
   • Video predictions: YoutubeLINK.

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.

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}}