Official pytorch implementation of Tsou et. al, "WLST: Weak Labels Guided Self-training for Weakly-supervised Domain Adaptation on 3D Object Detection", ICRA 2024.
In this paper, we propose a general weak labels guided self-training framework, WLST, designed for weakly-supervised domain adaptation (WDA) on 3D object detection. By incorporating autolabeler, which can generate 3D pseudo labels from 2D bounding boxes, into the existing self-training pipeline, our method is able to generate more robust and consistent pseudo labels that would benefit the training process on the target domain. Extensive experiments demonstrate the effectiveness, robustness, and detector-agnosticism of our WLST framework. Notably, it outperforms previous state-of-the-art methods on all evaluation tasks.
| Setting | Method | AP_BEV | AP_3D | Download | |
|---|---|---|---|---|---|
| PV-RCNN | - | Source Only | 60.32 | 21.66 | - |
| PV-RCNN | UDA | ST3D | 83.37 | 64.75 | - |
| PV-RCNN | WDA | ST3D (w/ SN) | 86.53 | 76.85 | - |
| PV-RCNN | WDA | WLST | 86.96 | 77.69 | model |
| PV-RCNN | - | Oracle | 90.85 | 83.00 | - |
| Setting | Method | AP_BEV | AP_3D | Download | |
|---|---|---|---|---|---|
| PV-RCNN | - | Source Only | 34.51 | 21.44 | - |
| PV-RCNN | UDA | ST3D | 36.38 | 22.99 | - |
| PV-RCNN | WDA | ST3D (w/ SN) | 36.65 | 23.66 | - |
| PV-RCNN | WDA | WLST | 39.54 | 24.46 | model |
| PV-RCNN | - | Oracle | 53.23 | 38.61 | - |
| Setting | Method | AP_BEV | AP_3D | Download | |
|---|---|---|---|---|---|
| PV-RCNN | - | Source Only | 69.26 | 39.17 | - |
| PV-RCNN | UDA | ST3D | 77.38 | 70.86 | - |
| PV-RCNN | WDA | ST3D (w/ SN) | 83.84 | 72.91 | - |
| PV-RCNN | WDA | WLST | 87.16 | 77.73 | model |
| PV-RCNN | - | Oracle | 90.85 | 83.00 | - |
| SECOND-IoU | - | Source Only | 51.84 | 17.92 | - |
| SECOND-IoU | UDA | ST3D | 75.94 | 54.13 | - |
| SECOND-IoU | WDA | ST3D (w/ SN) | 79.02 | 62.55 | - |
| SECOND-IoU | WDA | WLST | 80.67 | 64.65 | model |
| SECOND-IoU | - | Oracle | 83.29 | 73.45 | - |
Please refer to INSTALL.md.
Please refer to GETTING_STARTED.md.
Our code is released under the Apache 2.0 license.
Our code is heavily based on OpenPCDet v0.3 and ST3D. Thanks OpenPCDet Development Team for their awesome codebase.
If you find this project useful in your research, please consider cite:
@misc{tsou2023wlst,
title={WLST: Weak Labels Guided Self-training for Weakly-supervised Domain Adaptation on 3D Object Detection},
author={Tsung-Lin Tsou and Tsung-Han Wu and Winston H. Hsu},
journal={arXiv preprint arXiv:2310.03821},
year={2023},
}
@misc{openpcdet2020,
title={OpenPCDet: An Open-source Toolbox for 3D Object Detection from Point Clouds},
author={OpenPCDet Development Team},
howpublished={https://github.com/open-mmlab/OpenPCDet},
year={2020}
}