Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HyperDet: 3D Object Detection with Hyper 4D Radar Point Clouds

Yichun Xiao, Runwei Guan, Fangqiang Ding

TL;DR

HyperDet tackles the radar-only 3D detection gap by building a task-aware hyper 4D radar point cloud through spatio-temporal fusion, cross-sensor consensus validation, and a diffusion-based foreground enhancement guided by LiDAR supervision. The approach remains detector-agnostic and is compatible with standard LiDAR-oriented detectors, preserving radar-specific cues while densifying foreground geometry. Experiments on MAN TruckScenes show consistent gains over raw radar inputs with VoxelNeXt and CenterPoint, indicating effective input-level refinement that narrows the radar-LiDAR gap. This work demonstrates a practical path to deploy radar-enabled perception by leveraging existing LiDAR-oriented detectors with a refined radar front-end, particularly robust in adverse weather and cost-constrained scenarios.

Abstract

4D mmWave radar provides weather-robust, velocity-aware measurements and is more cost-effective than LiDAR. However, radar-only 3D detection still trails LiDAR-based systems because radar point clouds are sparse, irregular, and often corrupted by multipath noise, yielding weak and unstable geometry. We present HyperDet, a detector-agnostic radar-only 3D detection framework that constructs a task-aware hyper 4D radar point cloud for standard LiDAR-oriented detectors. HyperDet aggregates returns from multiple surround-view 4D radars over consecutive frames to improve coverage and density, then applies geometry-aware cross-sensor consensus validation with a lightweight self-consistency check outside overlap regions to suppress inconsistent returns. It further integrates a foreground-focused diffusion module with training-time mixed radar-LiDAR supervision to densify object structures while lifting radar attributes (e.g., Doppler, RCS); the model is distilled into a consistency model for single-step inference. On MAN TruckScenes, HyperDet consistently improves over raw radar inputs with VoxelNeXt and CenterPoint, partially narrowing the radar-LiDAR gap. These results show that input-level refinement enables radar to better leverage LiDAR-oriented detectors without architectural modifications.

HyperDet: 3D Object Detection with Hyper 4D Radar Point Clouds

TL;DR

HyperDet tackles the radar-only 3D detection gap by building a task-aware hyper 4D radar point cloud through spatio-temporal fusion, cross-sensor consensus validation, and a diffusion-based foreground enhancement guided by LiDAR supervision. The approach remains detector-agnostic and is compatible with standard LiDAR-oriented detectors, preserving radar-specific cues while densifying foreground geometry. Experiments on MAN TruckScenes show consistent gains over raw radar inputs with VoxelNeXt and CenterPoint, indicating effective input-level refinement that narrows the radar-LiDAR gap. This work demonstrates a practical path to deploy radar-enabled perception by leveraging existing LiDAR-oriented detectors with a refined radar front-end, particularly robust in adverse weather and cost-constrained scenarios.

Abstract

4D mmWave radar provides weather-robust, velocity-aware measurements and is more cost-effective than LiDAR. However, radar-only 3D detection still trails LiDAR-based systems because radar point clouds are sparse, irregular, and often corrupted by multipath noise, yielding weak and unstable geometry. We present HyperDet, a detector-agnostic radar-only 3D detection framework that constructs a task-aware hyper 4D radar point cloud for standard LiDAR-oriented detectors. HyperDet aggregates returns from multiple surround-view 4D radars over consecutive frames to improve coverage and density, then applies geometry-aware cross-sensor consensus validation with a lightweight self-consistency check outside overlap regions to suppress inconsistent returns. It further integrates a foreground-focused diffusion module with training-time mixed radar-LiDAR supervision to densify object structures while lifting radar attributes (e.g., Doppler, RCS); the model is distilled into a consistency model for single-step inference. On MAN TruckScenes, HyperDet consistently improves over raw radar inputs with VoxelNeXt and CenterPoint, partially narrowing the radar-LiDAR gap. These results show that input-level refinement enables radar to better leverage LiDAR-oriented detectors without architectural modifications.
Paper Structure (18 sections, 4 equations, 4 figures, 6 tables, 1 algorithm)

This paper contains 18 sections, 4 equations, 4 figures, 6 tables, 1 algorithm.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORKS
  3. LiDAR-based 3D Object Detection
  4. Radar-based 3D Object Detection
  5. Radar Point Cloud Enhancement
  6. Methodology
  7. Task Definition
  8. Overview
  9. Spatio-Temporal Fusion and Cross-Sensor Validation
  10. Diffusion-based Foreground Enhancement
  11. 3D Object Detector
  12. Experiment
  13. Experimental Setup
  14. Quantitative Results
  15. Ablation Studies
  16. ...and 3 more sections

Figures (4)

  • Figure 1: HyperDet enhances 4D radar input quality and improves 3D localization and orientation over raw radar. Camera annotations are for reference. Red boxes denote predictions; green boxes denote ground truth.
  • Figure 2: Overview of the proposed HyperDet pipeline. Raw multi-radar measurements are (i) temporally and spatially aggregated and validated via cross-sensor consensus, (ii) enhanced through diffusion-based foreground refinement, and (iii) processed by a LiDAR-compatible 3D detector.
  • Figure 3: Qualitative results of radar enhancement: raw radar input vs. diffusion-enhanced output vs. HyperDet detection results. Red boxes denote predictions; green boxes denote ground truth.
  • Figure 4: Comparison of diffusion enhancement under different supervision.