Description
3D point cloud segmentation needs lots of labeled training data, while 2D SAM already segments anything. SAM3D lifts Segment Anything's 2D masks into 3D scenes, segmenting point clouds without training.
It fuses SAM masks from multi-view RGB-D images into 3D; an ICCV 2023 workshop paper.
Training-free:Reuses SAM.
2D to 3D:Multi-view fusion.
Point clouds:Indoor scenes.
It fuses SAM masks from multi-view RGB-D images into 3D; an ICCV 2023 workshop paper.
Features
Training-free:Reuses SAM.
2D to 3D:Multi-view fusion.
Point clouds:Indoor scenes.
