Foundation Models for 3D Scenes: DINOv2 vs. CLIP for Instance Differentiation

This work moves beyond closed-set segmentation (Mask2Former) to open-set detection using SAM and Grounding DINO.