RaCFormer: Towards High-Quality 3D Object Detection via Query-based Radar-Camera Fusion (CVPR 2025)

Xiaomeng Chu, Jiajun Deng, Guoliang You, Yifan Duan, Houqiang Li, Yanyong Zhang

@inproceedings{chu2025racformer,
  title={RaCFormer: Towards High-Quality 3D Object Detection via Query-based Radar-Camera Fusion},
  author={Chu, Xiaomeng and Deng, Jiajun and You, Guoliang and Duan, Yifan and Li, Houqiang and Zhang, Yanyong},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={17081--17091},
  year={2025}
}

Overview

This repository is an official implementation of RaCFormer, an innovative query-based 3D object detection method through cross-perspective radar-camera fusion.

<img src="https://github.com/cxmomo/RaCFormer/raw/main/arch.jpg" alt="Dialogue_Teaser" width=100% >

Environment

Install PyTorch 2.0 + CUDA 11.8:

conda create -n racformer python=3.8
conda activate racformer
conda install pytorch==2.0.0 torchvision==0.15.0 pytorch-cuda=11.8 -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

Install turbojpeg and pillow-simd to speed up data loading (optional but important):

sudo apt-get update
sudo apt-get install -y libturbojpeg
pip install pyturbojpeg
pip uninstall pillow
pip install pillow-simd==9.0.0.post1

Compile CUDA extensions:

cd models/csrc
python setup.py build_ext --inplace

Prepare Dataset

1. Download nuScenes from https://www.nuscenes.org/nuscenes and put it in data/nuscenes.
2. Download the generated info files from Google Drive or generate the files by yourself using tools/gen_sweep_info.py.
3. Folder structure:

data/nuscenes
├── maps
├── nuscenes_infos_test_sweep.pkl
├── nuscenes_infos_train_sweep.pkl
├── samples
├── sweeps
├── v1.0-test
└── v1.0-trainval

Training

Download pretrained ResNet-50 and put it in directory pretrain/:

pretrain
├── cascade_mask_rcnn_r50_fpn_coco-20e_20e_nuim_20201009_124951-40963960.pth

Train RaCFormer with 8 GPUs:

torchrun --nproc_per_node 8 train.py --config configs/racformer_r50_nuimg_704x256_f8.py

Evaluation

Download the model weights.

Single-GPU evaluation:

export CUDA_VISIBLE_DEVICES=0
python val.py --config configs/racformer_r50_nuimg_704x256_f8.py --weights checkpoints/racformer_r50_f8.pth

Multi-GPU evaluation:

export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
torchrun --nproc_per_node 8 val.py --config configs/racformer_r50_nuimg_704x256_f8.py --weights checkpoints/racformer_r50_f8.pth

Acknowledgements

Many thanks to these excellent open-source projects:

• 3D Detection: SparseBEV, PETR v2, BEVFormer, BEVDet
• Codebase: MMDetection3D