Efficient 4D Radar Data Auto-labeling Method using LiDAR-based Object Detection Network

TL;DR

The paper tackles data scarcity for 4D radar–based 3D object detection under adverse weather by proposing an auto-labeling pipeline that uses a LiDAR-based detector trained on calibrated LPC to generate ground-truth ALs for the K-Radar train set. These ALs are refined via cross-frame IoU and used to train the 4D Radar Tensor Network with Height (RTNH), achieving detection performance close to an HL-trained baseline. The study shows that AL quality (driven by the chosen LODN) and label refinement materially affect RTNH results, with PVRCNN++-based ALs and refinement yielding the best outcomes, and that training on diverse weather data improves generalization. Overall, the method enables scalable expansion of 4D radar datasets like K-Radar, facilitating more robust 4D radar perception in autonomous driving.

Abstract

Focusing on the strength of 4D (4-Dimensional) radar, research about robust 3D object detection networks in adverse weather conditions has gained attention. To train such networks, datasets that contain large amounts of 4D radar data and ground truth labels are essential. However, the existing 4D radar datasets (e.g., K-Radar) lack sufficient sensor data and labels, which hinders the advancement in this research domain. Furthermore, enlarging the 4D radar datasets requires a time-consuming and expensive manual labeling process. To address these issues, we propose the auto-labeling method of 4D radar tensor (4DRT) in the K-Radar dataset. The proposed method initially trains a LiDAR-based object detection network (LODN) using calibrated LiDAR point cloud (LPC). The trained LODN then automatically generates ground truth labels (i.e., auto-labels, ALs) of the K-Radar train dataset without human intervention. The generated ALs are used to train the 4D radar-based object detection network (4DRODN), Radar Tensor Network with Height (RTNH). The experimental results demonstrate that RTNH trained with ALs has achieved a similar detection performance to the original RTNH which is trained with manually annotated ground truth labels, thereby verifying the effectiveness of the proposed auto-labeling method. All relevant codes will be soon available at the following GitHub project: https://github.com/kaist-avelab/K-Radar

Figures (3)

• Figure 1: Overview of 4DRT labeling process. 4DRT is not intuitive for human eyes to distinguish various objects in the data. In contrast, LPC can represent the detailed boundaries of objects, which can be utilized as clues to annotated ground truth labels. Therefore, calibration of 4DRT with LPC enables labeling of 4DRT by leveraging the detailed object representations from LPC.
• Figure 2: The overall process of the proposed auto-labeling method. The proposed method trains LODN (PVRCNN++ pvrcnn++) using calibrated LPC and HLs of the K-Radar test dataset. The method then generates ALs of the K-radar train dataset by saving inference outputs of trained LOND. The generated ALs are used to train 4DRODN (RTNHe-k-radar) to verify the effectiveness of the proposed method.
• Figure 3: Comparison of RTNH and RTNH-AL inference outputs in various weather conditions. In each example frame, HLs, inference outputs of RTNH, and those of RTNH-AL are presented on the 4DRT. The blue boxes represent the ground truth objects in HLs and the red boxes represent the detected objects of RTNH models..