tech π€ Radar learns traffic motion without expensive LiDAR π€
A research team developed IterFlow, a framework helping 4D radar estimate 3D motion in traffic scenes. This was done by Assistant Professor Zhao Na at SUTD. The method uses camera images and odometry instead of costly LiDAR supervision, showing results on the View-of-Delft dataset. IterFlow significantly outperformed CMFlow while being much lighter computationally. Future work will focus on overcoming the current fixed input point cloud size limitation for real-world use.