DyRAD: Radar Novel View Synthesis for Dynamic Driving Scenes

DyRAD reconstructs dynamic driving scenes from radar data by modeling them as static and moving point reflectors. It uses a physics-grounded renderer that separates scene structure from sensor noise by fixing the sensor's measurement spread. This allows for accurate synthesis of complete range, azimuth, and Doppler measurements, enabling novel view synthesis and zero-shot sensor transfer.

Open the full episode

More from AI Radio