BRIDGING THE GAP IN 3D DETECTION ACROSS DIFFERENT LIDAR ENVIRONMENTS

Summary from the TLDR AI newsletter

3D detectors often fail when used in unfamiliar settings, causing a huge drop in detection rates due to changes in tools, places, or weather. This project proposes a self-training system that adapts to different lidar types by creating accurate fake labels, MS3D++.