Code for the paper One Thing One Click: A Self-Training Approach for Weakly Supervised 3D Semantic Segmentation, CVPR 2021, and One Thing One Click++: Self-Training for Weakly Supervised 3D Scene Understanding.

Authors: Zhengzhe Liu, Xiaojuan Qi, Chi-Wing Fu

Update on April 16, 2023: One Thing One Click++ is on Arxiv, codes have been released.

Update on Jun 03, 2022: extend One-Thing-One-Click to instance segmentation. Please see the "instance_seg" folder.

Update on Jun 03, 2022: a new version on Scannet, achieves comparable performance on "cvpr2021_version". Please see the "scannet" folder.

Update on Dec 28, 2021: release the super-voxel generation, code and model on S3DIS. Please see the "s3dis" folder.

CVPR 2021 Version: Please see the "cvpr2021_version" folder.

This repo is built upon several repos, e.g., PointGrouop, SparseConvNet, spconv and ScanNet.

If you have any questions or suggestions about this repo, please feel free to contact me (liuzhengzhelzz@gmail.com).