AGO: Adaptive Grounding for Open World 3D Occupancy Prediction

Pretrained weights for AGO (ICCV 2025) on Occ3D-nuScenes. Paper: arXiv:2504.10117.

Weights

Each file contains the model state_dict. The CLIP weights are not included and are loaded from FeatUp at runtime.

File	Config	Known mIoU / Unknown mIoU / mIoU
ago_nuscenes_self_supervised.pth	ago_nuscenes_self_supervised.py	- / - / 19.16
ago_nuscenes_open_world_pretraining.pth	ago_nuscenes_open_world_pretraining.py	23.23 / 0.00 / 11.61
ago_nuscenes_open_world_pretraining.pth (zero-shot)	ago_nuscenes_open_world_pretraining.py	33.95 / 0.42 / 7.13
ago_nuscenes_open_world_finetuning.pth	ago_nuscenes_open_world_finetuning.py	36.88 / 8.78 / 14.40

Usage

hf download EdwardLeeLPZ/AGO --include "*.pth" --local-dir checkpoints
bash tools/dist_test.sh projects/configs/ago_nuscenes_self_supervised.py checkpoints/ago_nuscenes_self_supervised.pth 8 --tag self_supervised

See the code repository for installation and data preparation.

License

The weights are trained on 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}
}