IRGNN: Efficient Invariant Radar Graph Neural Network for Radar Point Cloud Object Detection

Xiao Guo; Wanke Xia; Lili Yang; Caicong Wu

cs.CVAug 14, 2026

IRGNN: Efficient Invariant Radar Graph Neural Network for Radar Point Cloud Object Detection

Authors: Xiao Guo, Wanke Xia, Lili Yang, Caicong Wu

Organizations: China Agricultural University · Tsinghua University

Abstract

Perception is a fundamental component of autonomous driving systems. While LiDAR-based methods have achieved remarkable progress in object detection, their reliability can degrade under adverse weather conditions. Radar point clouds provide a robust alternative due to their resilience to bad weather and low-illumination scenarios. However, radar point clouds are typically sparse, unordered, and less informative than LiDAR data, making it challenging to directly apply existing LiDAR-based perception methods. To address these challenges, we propose IRGNN, an Invariant Radar Graph Neural Network for radar point cloud object detection. IRGNN first reconstructs radar point clouds into graph representations using translation- and rotation-invariant feature designs, enabling robust modeling of sparse radar measurements. It then employs an improved message passing neural network (MPNN) with residual connections and a virtual node layer to enhance local feature propagation and global context modeling. Finally, task-specific heads are applied to the learned graph representations for object classification and bounding box prediction. Experimental results on the RadarScenes dataset show that IRGNN outperforms existing radar-based object detection methods and achieves competitive performance. In addition, IRGNN significantly reduces computational cost and memory usage during inference, demonstrating its effectiveness and practical potential for efficient radar-based perception in autonomous driving.

Figures & tables

Figure 1: Overview of IGRNN , mainly comprising Graph Construction and GNN .

Panel of Figure 2 — Figure 2: Schematic diagram of translation invariance: (a) no invariance (position changes after translation); (b) translation invariance (relative position does not change).

Panel of Figure 3 — Figure 3: Schematic diagram of rotation invariance: (a) RadarGNN’s [ 11 ] ; (b) Our EPPF.

Invariance	Node Feature	Edge Feature
Translation	$v_{x},v_{y},rcs,t_{idx},c$	$dx,dy$
Rotation	$\left\|v\right\|,rcs,t_{idx},c$	EPPF

Table 1: Node and edge feature for translation invariance and rotation invariance.

Model	mAP	Car	Pedestrian	Pedestrian Group	Two- Wheeler	Large Vehicle
DOG [ 19 ]	0.327	0.387	0.306	0.345	0.357	0.241
PointPillars [ 15 ]	0.392	0.148	0.148	0.279	0.405	0.549
PointNet++ [ 18 ]	0.459	0.531	0.336	0.515	0.522	0.390
RadarGNN [ 11 ]	0.567	0.709	0.277	0.560	0.588	0.701
IRGNN (Ours)	0.591	0.696	0.286	0.600	0.639	0.735

Table 2: Comparison results of mAP and class-wise AP by various detection methods.

Figure 4: Schematic diagram of different convolutional layers.

Layer	mAP	Car	Pedestrian	Pedestrian Group	Two- Wheeler	Large Vehicle
EGNN [ 20 ]	0.531	0.678	0.251	0.495	0.583	0.646
EGNNVel [ 20 ]	0.552	0.723	0.346	0.540	0.523	0.630
MPNN	0.567	0.709	0.277	0.560	0.588	0.701
MPNNVel	0.510	0.693	0.218	0.447	0.514	0.680

Table 3: Results of mAP and class-wise AP with different convolutional layers.

Figure 5: Training loss with different convolutional layers and improvements. All training processes have triggered the early stopping.

Model	mAP	Car	Pedestrian	Pedestrian Group	Two- Wheeler	Large Vehicle
RadarGNN [ 11 ]	0.229	0.229	0.118	0.267	0.337	0.196
RICONV++ [ 39 ]	0.230	0.258	0.102	0.256	0.339	0.193
IRGNN (Ours)	0.238	0.259	0.133	0.332	0.277	0.190

Table 4: Results of mAP and class-wise AP with different rotation invariance pipelines.

Residual Blocks	Virtual Node Layer	mAP	Car	Pedestrian	Pedestrian Group	Two- Wheeler	Large Vehicle
$\times$	$\times$	0.567	0.709	0.277	0.560	0.588	0.701
✓	$\times$	0.578	0.727	0.242	0.570	0.638	0.711
✓	✓	0.591	0.696	0.286	0.600	0.639	0.735

Table 5: Results of mAP and class-wise AP in ablation studies.

Model	Raw Box Cost (s)	Extraction Cost (s)	NMS Cost (s)	VRAM Usage (GB)
Baseline	224.23	131.43	21.61	8.11
+Vectorization	225.34	30.19	20.94	8.10
+Intermediate Release	176.28	32.29	22.73	3.05

Table 6: Comparisons of post-process time cost and VRAM usage. Time cost refers to the total inference time for the entire validation set. VRAM usage represents the consumption during the raw bounding box generation stage.

3D Object Detection

3D object detection identifies objects and estimates their locations, sizes, and orientations from images, point clouds, radar, or fused sensors.

See all 181 papers

RADE-Net: Robust Attention Network for Radar-Only Object Detection in Adverse Weather
Feb 23, 2026Christof Leitgeb, Thomas Puchleitner, Max Peter Ronecker +13D Object Detection mmWave Radar Sensing
Infineon Technologies AG, Institute of Visual Computing, Graz University of Technology, Graz, Austria · Infineon Technologies AG, Graz, Austria · Virtual Vehicle Research GmbH, Institute of Visual Computing, Graz University of Technology, Graz, Austria +1
SD4R: Sparse-to-Dense Learning for 3D Object Detection with 4D Radar
Feb 24, 2026Xiaokai Bai, Jiahao Cheng, Songkai Wang +53D Object Detection Point Cloud Learning
College of Information Science and Electronic Engineering, Zhejiang University, Hangzhou 310027, China · School of Automotive Studies, Tongji University, Shanghai 201804, China · Ningbo Global Innovation Center, Zhejiang University, Ningbo 315100, China
4D-RaDiff: Latent Point Diffusion for 4D Radar Point Cloud Generation
Dec 16, 2025Jimmie Kwok, Holger Caesar, Andras PalffySynthetic Data Augmentation Point Cloud Learning
Delft University of Technology, Netherlands · Perciv AI, Netherlands

Feb 23, 2026J/K move · Enter open · S save

RADE-Net: Robust Attention Network for Radar-Only Object Detection in Adverse Weather

Christof Leitgeb, Thomas Puchleitner, Max Peter Ronecker, Daniel Watzenig

Infineon Technologies AG, Institute of Visual Computing, Graz University of Technology, Graz, Austria · Infineon Technologies AG, Graz, Austria · Virtual Vehicle Research GmbH, Institute of Visual Computing, Graz University of Technology, Graz, Austria+1

3D Object Detection mmWave Radar Sensing

Open paper AskPDF arXiv

Automotive perception systems are obligated to meet high requirements. While optical sensors such as Camera and Lidar struggle in adverse weather conditions, Radar provides a more robust perception performance, effectively penetrating fog, rain, and snow. Since full Radar tensors have large data sizes and very few datasets provide them, most Radar-based approaches work with sparse point clouds or 2D projections, which can result in information loss. Additionally, deep learning methods show potential to extract richer and more dense features from low level Radar data and therefore significantly increase the perception performance. Therefore, we propose a 3D projection method for fast-Fourier-transformed 4D Range-Azimuth-Doppler-Elevation (RADE) tensors. Our method preserves rich Doppler and Elevation features while reducing the required data size for a single frame by 91.9% compared to a full tensor, thus achieving higher training and inference speed as well as lower model complexity. We introduce RADE-Net, a lightweight model tailored to 3D projections of the RADE tensor. The backbone enables exploitation of low-level and high-level cues of Radar tensors with spatial and channel-attention. The decoupled detection heads predict object center-points directly in the Range-Azimuth domain and regress rotated 3D bounding boxes from rich feature maps in the cartesian scene. We evaluate the model on scenes with multiple different road users and under various weather conditions on the large-scale K-Radar dataset and achieve a 16.7% improvement compared to their baseline, as well as 6.5% improvement over current Radar-only models. Additionally, we outperform several Lidar approaches in scenarios with adverse weather conditions. The code is available under https://github.com/chr-is-tof/RADE-Net.

Abstract

3D Object Detection

Explore similar work

RADE-Net: Robust Attention Network for Radar-Only Object Detection in Adverse Weather

SD4R: Sparse-to-Dense Learning for 3D Object Detection with 4D Radar

4D-RaDiff: Latent Point Diffusion for 4D Radar Point Cloud Generation