PolarBEVDet

This is the official PyTorch implementation for our paper:

PolarBEVDet: Exploring Polar Representation for Multi-View 3D Object Detection in Bird's-Eye-View

Model Zoo

Setting	Pretrain	NDS	MAP	Weights
r50_704x256_24e	ImageNet	53.0	43.2	gdrive
r50_704x256_60e	ImageNet	55.3	45.0	gdrive
r50_704x256_nuImg_60e	nuImg	56.7	46.9	gdrive

Environment

conda create -n polarbevdet python=3.9
conda activate polarbevdet
conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia

Install other dependencies:

pip install openmim
mim install mmcv-full==1.6.0
mim install mmdet==2.28.2
mim install mmsegmentation==0.30.0
mim install mmdet3d==1.0.0rc6
pip install setuptools==59.5.0
pip install numpy==1.23.5
pip install pillow==8.4.0

Compile CUDA extensions:

cd projects
python setup.py develop

Prepare Dataset

1. Download nuScenes from https://www.nuscenes.org/nuscenes and put it in data/nuscenes.
2. Generate info files by:

python tools/create_data_bevdet.py

3. Folder structure:

data/nuscenes
├── maps
├── nuscenes_infos_test.pkl
├── nuscenes_infos_train.pkl
├── nuscenes_infos_val.pkl
├── samples
├── sweeps
├── v1.0-test
└── v1.0-trainval

Training

Train PolarBEVDet with 4 GPUs:

bash tools/dist_train.sh projects/configs/polarbevdet/r50_704x256_24e.py 4 --work-dir work_dirs/polarbevdet/r50_704x256_24e

Evaluation

python tools/swap_ema_and_non_ema.py work_dirs/polarbevdet/r50_704x256_24e/iter_21096.pth

bash tools/dist_test.sh projects/configs/polarbevdet/r50_704x256_24e.py work_dirs/polarbevdet/r50_704x256_24e/iter_21096_ema.pth 4 --eval map

Acknowledgements

Many thanks to these excellent open-source projects:

• 3D Detection:BEVDet, SOLOFusion, StreamPETR
• Codebase: MMDetection3D

Bibtex

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

@article{yu2024polarbevdet,
  title={PolarBEVDet: Exploring Polar Representation for Multi-View 3D Object Detection in Bird's-Eye-View},
  author={Yu, Zichen and Liu, Quanli and Wang, Wei and Zhang, Liyong and Zhao, Xiaoguang},
  journal={arXiv preprint arXiv:2408.16200},
  year={2024}
}