3D Object Discovery

3D object discovery focuses on automatically identifying and isolating individual objects within complex 3D scenes captured from multiple images or point clouds, without relying on manual annotations. Current research emphasizes class-agnostic approaches, using techniques like neural radiance fields (NeRFs) and 3D object detection networks trained with geometric cues and/or self-supervised learning to discover objects even when their classes are unknown. These advancements are significant for applications such as automated 3D scene understanding, digital content creation, and robotics, enabling more efficient and robust processing of 3D data.