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

2025-12-16

AI Summary

The scarcity of annotated 4D automotive radar point clouds severely limits perception performance. Method: This paper introduces diffusion models to sparse radar point cloud generation for the first time, proposing a latent-space-based conditional diffusion framework. It establishes dual-level (object- and scene-level) conditioning mechanisms to jointly model radar-specific geometric structures, Doppler motion characteristics, and physical priors within a compact latent point cloud space. The framework supports high-fidelity cross-modal generation (LiDAR → Radar) driven either by label-free bounding boxes or LiDAR data. Results: Generated radar point clouds significantly improve detection performance; only 10% of real annotations are required to match fully supervised performance, reducing annotation cost by 90%. This work establishes a scalable, low-resource paradigm for radar perception.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Automotive radar has shown promising developments in environment perception due to its cost-effectiveness and robustness in adverse weather conditions. However, the limited availability of annotated radar data poses a significant challenge for advancing radar-based perception systems. To address this limitation, we propose a novel framework to generate 4D radar point clouds for training and evaluating object detectors. Unlike image-based diffusion, our method is designed to consider the sparsity and unique characteristics of radar point clouds by applying diffusion to a latent point cloud representation. Within this latent space, generation is controlled via conditioning at either the object or scene level. The proposed 4D-RaDiff converts unlabeled bounding boxes into high-quality radar annotations and transforms existing LiDAR point cloud data into realistic radar scenes. Experiments demonstrate that incorporating synthetic radar data of 4D-RaDiff as data augmentation method during training consistently improves object detection performance compared to training on real data only. In addition, pre-training on our synthetic data reduces the amount of required annotated radar data by up to 90% while achieving comparable object detection performance.

Problem

• Generates synthetic 4D radar point clouds for training
• Addresses limited annotated radar data availability
• Improves object detection with data augmentation

Innovation

• Latent diffusion for sparse 4D radar point clouds
• Conditioning generation at object or scene level
• Converts unlabeled boxes and LiDAR into radar data