Awesome

DualBEV

This repository is an official implementation of DualBEV.

If you find this work interesting, please do not hesitate to give a ⭐!

News

2024.07.01 DualBEV is accepted to ECCV 2024. 🎉
2024.07.01 Code of DualBEV is released. 🚀
2024.04.22 Blog in Chinese Zhihu.🔥
2024.03.08 Paper of DualBEV available in arxiv.

Main Results

Config	Frames	mAP	NDS	Model	Log
DualBEV-R50-CBGS	<p align="center">1	36.8	43.4	google / baidu	google / baidu
DualBEV-R50-4D-CBGS	<p align="center">2	38.6	50.5	google / baidu	google / baidu

Get Started

Installation and Data Preparation

Please refer to getting_started.md for installing DualBEV as mmdetection3d. Docker is recommended for environment preparation.
Prepare nuScenes dataset as introduced in nuscenes_det.md and create the pkl for DualBEV by running:

python tools/create_data_bevdet.py

Download nuScenes-lidarseg from nuScenes official site and put it under data/nuscenes/. Create depth and semantic labels from point cloud by running:

python tools/generate_point_label.py

Train and Test

Single-Frame:

bash tools/dist_train.sh configs/dualbev/dualbev-r50-cbgs.py 8 --no-validate
bash tools/dist_test.sh configs/dualbev/dualbev-r50-cbgs.py  work_dirs/dualbev-r50-cbgs/epoch_20_ema.pth 8 --eval mAP

Multi-Frame:

bash tools/dist_train.sh configs/dualbev/dualbev4d-r50-cbgs.py 8 --no-validate
bash tools/dist_test.sh configs/dualbev/dualbev4d-r50-cbgs.py  work_dirs/dualbev4d-r50-cbgs/epoch_20_ema.pth 8 --no-aavt --eval mAP

Acknowledgement

This project is not possible without multiple great open-sourced code bases. We list some notable examples below.

Bibtex

If this work is helpful for your research, please consider citing the following BibTeX entry.

@article{li2024dualbev,
      title={DualBEV: CNN is All You Need in View Transformation}, 
      author={Peidong Li and Wancheng Shen and Qihao Huang and Dixiao Cui},
      journal={arXiv preprint arXiv:2403.05402},
      year={2024},
}