Skip to content

SAM-Graph

A graph-cut method for 3D instance segmentation

Visit official site

What SAM-Graph is

The method is described as SAM-guided Graph Cut for 3D Instance Segmentation, a deep learning approach that uses 3D geometry and multi-view images for segmentation.

What you can do with it

  • Segment surroundings to track vehicles and pedestrians in driving
  • Segment indoor scenes for robot path planning
  • Segment real scenes for augmented reality placement

Key features

  • 3D-to-2D query framework for scene generalisation
  • Superpoint graph built from a graph-cut problem
  • Validated on ScanNet, ScanNet++ and KITTI-360

Pros & cons

✓ What we like

  • Combines 3D geometry with multi-view image data
  • Reports robust segmentation across scene types

! What to watch out for

  • The workflow needs 2D segmentation output first
  • Training relies on pseudo 3D labels

FAQ

What data does it use?

The material names 3D geometry and multi-view images.

Where was it validated?

The listed datasets are ScanNet, ScanNet++ and KITTI-360.

Last reviewed: 2026-11-15

More AI text detection tools

View all →

How we review