EdwardLeeLPZ/AGO-Occ3D-nuScenes-pseudo-labels

AGO Occ3D-nuScenes Pseudo Labels

3D occupancy pseudo labels used to train AGO (ICCV 2025). The 2D pseudo labels are generated with Grounded SAM and lifted to the Occ3D-nuScenes voxel grid. Paper: arXiv:2504.10117.

Content

occ3d_pseudo_labels.tar contains 34,149 frames from the 850 scenes of nuScenes v1.0-trainval:

pseudo_labels
└── [scene_name]/[sample_token]/labels.npz

Each labels.npz stores semantics, an int8 array of shape (1, 200, 200, 16) on the Occ3D-nuScenes grid. Values 0-44 index the prompts in LABEL_PROMPT of label_space.py, 45 is free space and -1 is ignored.

Usage

hf download EdwardLeeLPZ/AGO-Occ3D-nuScenes-pseudo-labels occ3d_pseudo_labels.tar --repo-type dataset --local-dir .
tar xf occ3d_pseudo_labels.tar -C data/nuscenes/occ3d

License

The pseudo labels are derived from nuScenes and subject to the nuScenes terms of use.

Citation

@inproceedings{li2025ago,
  title={AGO: Adaptive Grounding for Open World 3D Occupancy Prediction},
  author={Li, Peizheng and Ding, Shuxiao and Zhou, You and Zhang, Qingwen and Inak, Onat and Triess, Larissa and Hanselmann, Niklas and Cordts, Marius and Zell, Andreas},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  pages={8645--8655},
  year={2025}
}