One Class One Click (OCOC): Quasi Scene-level Weakly Supervised Point Cloud Semantic Segmentation with Active Learning
Official code for the paper One Class One Click: Quasi Scene-level Weakly Supervised Point Cloud Semantic Segmentation with Active Learning. OCOC is a active weakly supervised learning point cloud semantic segmentation method (paper link).
- Create conda environment
conda create -n ococ python=3.8
conda activate ococ
- Install the package using
pip install -r requirement.txt
- Please follow the KPConv to compile the C++ extension modules in
cpp_wrappers
python train_H3D.py
For validation set
python test_model -s "H3D/log_name" -m val
For test set
python test_model -s "H3D/log_name" -m test
python cal_metrics.py
If you find our work useful in your research, please consider citing:
@article{WANG202389,
title = {One Class One Click: Quasi scene-level weakly supervised point cloud semantic segmentation with active learning},
journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
volume = {204},
pages = {89-104},
year = {2023},
issn = {0924-2716},
doi = {https://doi.org/10.1016/j.isprsjprs.2023.09.002},
url = {https://www.sciencedirect.com/science/article/pii/S0924271623002344},
author = {Puzuo Wang and Wei Yao and Jie Shao},
}
Our code uses the KPConv as the backbone network.
Our code is released under MIT License (see LICENSE file for details).