UNION: Unsupervised 3D Object Detection using Appearance-based Pseudo-Classes [NeurIPS'24]

[arXiv] [BibTeX]

• [2024-10-31] Camera-ready release on arXiv (v2)!
• [2024-09-25] UNION has been accepted for NeurIPS'24!
• [2024-05-24] Paper release on arXiv (v1)!

Installation

Coming soon!

Generate pseudo-labels

Coming soon!

Results on nuScenes [class-agnostic detection]

Class-agnostic 3D object detection performance on the nuScenes validation split (150 scenes). For each object discovery method, the detector CenterPoint has been trained with the method's generated pseudo-bounding boxes on the nuScenes training split (700 scenes). The AAE is set to 1.0 by default for all methods. L and C stand for LiDAR and camera, respectively.

Method	Conference	Labels	Self-Training	AP ↑	NDS ↑	ATE ↓	ASE ↓	AOE ↓	AVE ↓
HDBSCAN	JOSS'17	L	❎	13.8	15.9	0.574	0.522	1.601	1.531
OYSTER	CVPR'23	L	☑️	9.1	11.5	0.784	0.521	1.514	-
LISO	ECCV'24	L	☑️	10.9	13.9	0.750	0.409	1.062	-
UNION (ours)	NeurIPS'24	L+C	❎	38.4	31.2	0.589	0.497	0.874	0.836

Results on nuScenes [multi-class detection]

Coming soon!

Citation Information

If UNION is useful to your research, please kindly recognize our contributions by citing our paper.

@article{lentsch2024union,
  title={{UNION}: Unsupervised {3D} Object Detection using Object Appearance-based Pseudo-Classes},
  author={Lentsch, Ted and Caesar, Holger and Gavrila, Dariu M},
  journal={Advances in Neural Information Processing Systems},
  year={2024}
}