---
pretty_name: nuScenes-UNION-labels
license: cc-by-nc-sa-4.0
task_categories:
- object-detection
tags:
- mmdetection3d
- nuscenes
- pseudo-labels
- robotics
- union
- unsupervised
language:
- en
arxiv: 2405.15688
papers:
- title: "UNION: Unsupervised 3D Object Detection using Appearance-based Pseudo-Classes"
  url: https://arxiv.org/abs/2405.15688
  conference: NeurIPS 2024
citations:
- type: arxiv
  url: https://arxiv.org/abs/2405.15688
- type: google_scholar
  url: https://scholar.google.com/citations?view_op=view_citation&hl=en&user=54NWkMoAAAAJ&citation_for_view=54NWkMoAAAAJ:roLk4NBRz8UC
- type: bibtex
  content: |
    @inproceedings{lentsch2024union,
      title={{UNION}: Unsupervised {3D} Object Detection using Object Appearance-based Pseudo-Classes},
      author={Lentsch, Ted and Caesar, Holger and Gavrila, Dariu M},
      booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
      year={2024}
    }
---


# UNION: Unsupervised 3D Object Detection using Appearance-based Pseudo-Classes [NeurIPS 2024]

[[`arXiv`](https://arxiv.org/abs/2405.15688)] [[`GitHub`](https://github.com/TedLentsch/UNION)] [[`BibTeX`](#citation-information)]

Hugging Face repository for prebuilt .pkl annotation files created by [UNION](https://github.com/TedLentsch/UNION) on the [nuScenes](https://arxiv.org/abs/1903.11027) dataset.

This repo is **annotations-only** (no sensor data) so you can reproduce UNION results quickly without regenerating labels.

+ 🔗 **Code:** https://github.com/TedLentsch/UNION  
+ ⬇️ **nuScenes download:** https://www.nuscenes.org/nuscenes  
+ 🛡️ **License:** `CC BY-NC-SA 4.0` *(inherits from nuScenes — non-commercial, share-alike)*


---
## Pipeline Overview
![](figures/figure1.jpg)


---
## Contents

This repo contains three top-level folders, namely `standard-labels`, `class-agnostic-labels`, and `multi-class-003-labels`.
The folder `standard-labels` contains the MMDetection3D infos from the ground truth annotations for the standard dataset split and classes.
The folders `class-agnostic-labels` and `multi-class-003-labels` contain the ground truth labels and pseudo-labels for class agnostic and multi-class (vehicle, pedestrian, cyclist) 3D object detection, respectively.


---
## Citation Information

```bibtex
@inproceedings{caesar2020nuscenes,
  title={{nuScenes}: A multimodal dataset for autonomous driving},
  author={Caesar, Holger and Bankiti, Varun and Lang, Alex H and Vora, Sourabh and Liong, Venice Erin and Xu, Qiang and Krishnan, Anush and Pan, Yu and Baldan, Giancarlo and Beijbom, Oscar},
  booktitle={Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages={11621--11631},
  year={2020}
}

@misc{mmdet3d2020,
  title={{MMDetection3D: OpenMMLab} next-generation platform for general {3D} object detection},
  author={MMDetection3D Contributors},
  howpublished={\url{https://github.com/open-mmlab/mmdetection3d}},
  year={2020}
}

@inproceedings{lentsch2024union,
  title={{UNION}: Unsupervised {3D} Object Detection using Object Appearance-based Pseudo-Classes},
  author={Lentsch, Ted and Caesar, Holger and Gavrila, Dariu M},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  pages={22028--22046},
  volume={37},
  year={2024}
}
````

---

## Questions & issues

* Problems or feature requests? Please open an issue in **[UNION GitHub Issues](https://github.com/TedLentsch/UNION/issues)**.

---

# License

This repository is distributed under **CC BY-NC-SA 4.0** to comply with the nuScenes license.
**Non-commercial use only**, and derivatives must be shared under the same license.

+ You must also comply with the original **nuScenes** license when using these annotations.
+ If you use these files, please cite **nuScenes**, **MMDetection3D**, and **UNION**.

Note: UNION code base itself is released under **Apache-2.0 license**, which allows for commercial use!

[![License: CC BY-NC-SA 4.0](https://img.shields.io/badge/License-CC%20BY--NC--SA%204.0-lightgrey.svg)](https://creativecommons.org/licenses/by-nc-sa/4.0/)