/RCDPT

Primary LanguagePythonMIT LicenseMIT

RCDPT

Radar-Camera Fusion Dense Prediction Transformer

Official implementation of "RCDPT: Radar-Camera Fusion Dense Prediction Transformer" (https://arxiv.org/abs/2211.02432), an accepted paper of ICASSP2023.

Dependency

Please check Dockerfile for environment settings and python packages

Or you could directly use the pre-built docker image with tag 'lochenchou/det:mde' from docker hub.

Usage

Generating Multi-channel Enhanced Radar

For generating Multi-channel Enhanced Radar (MER), which is the radar format used in the experiment, please follow the instructions of RC-PDA (https://github.com/longyunf/rc-pda).

Generating Sparse Lidar depthmap.

Please follow the steps in gen_interpolation.py in DORN_radar repo (https://github.com/lochenchou/DORN_radar) to generate 5 frames sparse lidar as the golden ground truth to evaluate with during the evaluation step.

Train baseline model and proposed model on nuScenes

Directly call 'train.py' with dataset paths for training the baseline model and the proposed model on nuScenes.

Citation

If you find this work useful in your research, please consider citing:

@inproceedings{RCDPT,
  author={Lo, Chen-Chou and Vandewalle, Patrick},
  booktitle={Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2023)}, 
  title={RCDPT: Radar-Camera fusion Dense Prediction Transformer}, 
  year={2023},
}