RadarGen: Diffusion Model Generates Automotive Radar Point Clouds from Camera Images

A new diffusion model named RadarGen has been developed by researchers to create realistic automotive radar point clouds using images from multiple cameras. This model presents radar data in a bird's-eye view, capturing both spatial structure and attributes such as radar cross section (RCS) and Doppler information. A simple recovery process reconstructs point clouds from these generated maps. To ensure the generated output aligns with the visual environment, RadarGen uses depth, semantic, and motion cues from pretrained foundation models, steering the generation towards realistic radar patterns. This image-based conditioning enhances compatibility with various visual datasets and simulation frameworks, paving the way for scalable multimodal generative simulations. Evaluations on extensive driving datasets indicate RadarGen's strong performance. The research paper can be found on arXiv with the identifier 2512.17897.