This is the official implementation of the paper: Class-Aware PillarMix: Can Mixed Sample Data Augmentation Enhance 3D Object Detection with Radar Point Clouds? (IROS 20205). The paper can be found here: https://arxiv.org/abs/2503.02687. The code is built on top of the mmdet3d framework and aim to reproduce the results on K-radar dataset, but also compatible with other datasets in mmdet3d.
We propose Class-Aware PillarMix (CAPMix), a novel MSDA approach that applies MixUp at the pillar level in 3D point clouds, guided by class labels. Unlike methods that rely a single mix ratio to the entire sample, CAPMix assigns an independent ratio to each pillar, boosting sample diversity. This class-aware mixing retains critical details and enriches each sample with new information, ultimately generating more diverse training data.
This project is actively being developed. Expect frequent changes to the code structure and possible unstable behavior or bugs.
Please follow the installation instructions of mmdet3d.
After that, put the code in this repo to the mmdetection3d/projects/ folder.
We use the K-Radar dataset for training and evaluation. Please get access here. The dataset and evaluation metric is reimplemented in mmdet3d based on offcial K-radar implementation. Please check the details here. For effeiciency, we use the CFAR-point cloud processed from the raw radar data. The processing details can be found in gen_cfar.py. Please modify the paths based on your own settings.
We implement the augmentation methods in LaserMix-style in mmdet3d. So in principle, it is also compatible with other datasets in mmdet3d. Feel free to try it on other datasets.😄
We provide the config files for centerpoint here.
To train a model with CAPMix, run the following command:
python tools/train.py configs/centerpoint_kradar_capmix.pyIf you find this work helpful, please kindly consider citing our paper:
@article{zhang2025class,
title={Class-Aware PillarMix: Can Mixed Sample Data Augmentation Enhance 3D Object Detection with Radar Point Clouds?},
author={Zhang, Miao and Abdulatif, Sherif and Loesch, Benedikt and Altmann, Marco and Yang, Bin},
journal={arXiv preprint arXiv:2503.02687},
year={2025}
}
Class-Aware PillarMix is open-sourced under the AGPL-3.0 license. See the LICENSE file for details.
For a list of other open source components included in Class-Aware PillarMix, see the file 3rd-party-licenses.txt.
This work is developed based on the MMDetection3D codebase.
We acknowledge the use of the following public resources during the course of this work: K-Radar dataset, LaserMix, PolarMix, SSDA3D.