Annotator for LiDAR Semantic Segmentation

Annotator: An Generic Active Learning Baseline for LiDAR Semantic Segmentation

Binhui Xie, Shuang Li, Qingju Guo, Chi Harold Liu and Xinjing Cheng

Highlight

🌈 we present a voxel-centric online active learning baseline that efficiently reduces the labeling cost of enormous point clouds and effectively facilitates learning with a limited budget.
⚖️ we introduce a novel label acquisition strategy, voxel confusion degree (VCD), that requires 1000× fewer annotations while reaching a close segmentation performance to that of the fully supervised counterpart.
🚀 Annotator is generally applicable and works for different network architectures (e.g., MinkNet, SPVCNN, etc.), in distribution or out of distribution setting (i.e., AL, ASFDA, and ADA), and simulation-to-real (SynLiDAR→SemanticKITTI/SemanticPOSS) and real-to-real (SemanticKITTI→nuScenes and nuScenes→SemanticKITTI) scenarios with consistent performance gains

Usage

Prerequisites

Please see INSTALL.md.

Data Preparation

Please see DATA.md

Training and Evaluation

Please see TRAIN_VAL.md

Citation

If you find this project useful in your research, please consider citing:

@inproceedings{xie2023annotator,
 author = {Binhui Xie, Shuang Li, Qingju Guo, Chi Harold Liu, Xinjing Cheng},
 booktitle = {Advances in Neural Information Processing Systems},
 title = {Annotator: An Generic Active Learning Baseline for LiDAR Semantic Segmentation},
 year = {2023}
}

Acknowledgements

This project is based on the following projects: OpenPCDet, PCSeg, LaserMix and SynLiDAR. We thank their authors for making the source code publicly available.

Contact

For help and issues associated with Annotator, or reporting a bug, please open a [GitHub Issues], or feel free to contact binhuixie@bit.edu.cn.

BIT-DA/Annotator