Forward-Backward View Transformation for Vision-Centric AV Perception

Paper (FB-BEV) | Paper (FB-OCC) | Intro Video

FB-BEV and FB-OCC are a family of vision-centric 3D object detection and occupancy prediction methods based on forward-backward view transformation.

News

[2023/8/01] FB-BEV was accepted to ICCV 2023.
🏆 [2023/6/16] FB-OCC wins both Outstanding Champion and Innovation Award in Autonomous Driving Challenge in conjunction with CVPR 2023 End-to-End Autonomous Driving Workshop and Vision-Centric Autonomous Driving Workshop.

Getting Started

Model Zoo

Backbone	Method	Lr Schd	IoU	Config	Download
R50	FB-OCC	20ep	39.1	config	model

More model weights will be released later.

License

This work is made available under the Nvidia Source Code License-NC. Click here to view a copy of this license.

The pre-trained models are shared under CC-BY-NC-SA-4.0. If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original.

For business inquiries, please visit our website and submit the form: NVIDIA Research Licensing.

Citation

If this work is helpful for your research, please consider citing:

@inproceedings{li2023fbbev,
  title={{FB-BEV}: {BEV} Representation from Forward-Backward View Transformations},
  author={Li, Zhiqi and Yu, Zhiding and Wang, Wenhai and Anandkumar, Anima and Lu, Tong and Alvarez, Jose M},
  booktitle={IEEE/CVF International Conference on Computer Vision (ICCV)},
  year={2023}
}

@article{li2023fbocc,
  title={{FB-OCC}: {3D} Occupancy Prediction based on Forward-Backward View Transformation},
  author={Li, Zhiqi and Yu, Zhiding and Austin, David and Fang, Mingsheng and Lan, Shiyi and Kautz, Jan and Alvarez, Jose M},
  journal={arXiv:2307.01492},
  year={2023}
}

Acknowledgement

Many thanks to these excellent open source projects:

BEVFormer, BEVDet, Occ3D, OpenOccupancy, SoloFusion

NVlabs / FB-BEV

readme