SRCN3D: Sparse R-CNN 3D Surround-View Cameras 3D Object Detection and Tracking for Autonomous Driving

This repo is the official implementations of SRCN3D (https://arxiv.org/abs/2206.14451). Our implementation is based on MMdetection3D.

Preparation

Please install the latest version of mmdet3d (https://github.com/open-mmlab/mmdetection3d) according to Open-MMLab guidelines. Give a soft link of mmdetection3d with

• Reference Environments
  Linux(Ubuntu18.04LTS), Python==3.7.13, CUDA == 11.3, pytorch == 1.10.2,torchvision == 0.11.3 ,mmdet3d == 0.17.2

ln -s /path/to/mmdetection3d {/path/to/SRCN3D}/

Data

1. Follow the mmdet3d to process the nuScenes dataset (https://github.com/open-mmlab/mmdetection3d/blob/master/docs/en/data_preparation.md).

Train

1. Downloads the [pretrained backbone weights] from DETR3D repository (https://drive.google.com/drive/folders/1h5bDg7Oh9hKvkFL-dRhu5-ahrEp2lRNN?usp=sharing) to ckpts/

2. For example, to train a basic version of SRCN3D on 2 GPUs, please use

bash tools/dist_train.sh projects/configs/srcn3d/srcn3d_res101_roi7_nusc.py 2

Pretrained models

1. Download the weights accordingly.

Backbone	mAP	NDS	Download
SRCN3D, ResNet101 w/ DCN	33.7	42.8	model | log
SRCN3D, V2-99	39.6	47.5	model | log

2. for a validation and test submission, use
   tools/dist_test.sh path/to/config.py /path/to/ckpt 1 --eval=bbox

Bibtex

If you find this repo useful for your research, please consider citing the papers

@inproceedings{SRCN3D,
  doi = {10.48550/ARXIV.2206.14451},
  url = {https://arxiv.org/abs/2206.14451},
  author = {Shi, Yining and Shen, Jingyan and Sun, Yifan and Wang, Yunlong and Li, Jiaxin and Sun, Shiqi and Jiang, Kun and Yang, Diange},
  title = {SRCN3D: Sparse R-CNN 3D Surround-View Camera Object Detection and Tracking for Autonomous Driving},
  journal={arXiv preprint arXiv:2206.14451},
  publisher = {arXiv},
  year = {2022},
}

News

• [2022/6/27]: We release an initial version of SRCN3D.

Acknowledgement

Thanks to prior excellent open source projects:

• DETR3D
• mmdet3d

Contact

The repository is still in an early stage, if you have any questions, feel free to open an issue or contact us at syn21@mails.tsinghua.edu.cn.