ES-EOT-real-nuscenes

If you use the code, please cite our paper:

@article{deng20243d,
  title={3D Extended Object Tracking by Fusing Roadside Sparse Radar Point Clouds and Pixel Keypoints},
  author={Jiayin Deng and Zhiqun Hu and Yuxuan Xia and Zhaoming Lu and Xiangming Wen},
  journal={arXiv preprint arXiv:2404.17903},
  year={2024},
  doi={2404.17903}
}

Installation

• Clone the repository and cd to it.

  git clone https://github.com/RadarCameraFusionTeam-BUPT/ES-EOT-real-nuscenes.git
  cd ES-EOT

• Install dependencies according to the installation part of github page.

Usage

Note: The detections of CRN and HVDetFusion are saved in json files, while the realtime detection is not released here.

• Run the main file. (The result is written in a .npy file in the same folder as main.py)

  cd ES-EOT
  python main.py

• Calculate the ATE and ASE values.

  python calculate_matrics.py

• Show the tracking results in an animation.

  python ShowRes.py

Show

• Show BEV detections in a .jpg picture

  cd show
  python show_BEV_track.py

Keypoints detection using pre-trained model

Note: Keypoints detections are stored in the data/turn_left/kps_json/. However, if you wish to reuse the pre-trained model, follow these steps:

1. Download the pre-trained model

• Obtain the pre-trained model weights from model link, and move it into the assets folder.

2. Run Keypoints Detection

• Once you have the pre-trained model weights and dependencies installed, you can run the key points detection script

  python predict.py data/turn_left/images --model assets/best.pt --render

  The results are written into the output folder.