SAM2Point: Segment Any 3D as Videos in Zero-shot and Promptable Manners

TL;DR

This work introduces Sam2Point, a zero-shot, promptable 3D segmentation framework that repurposes SAM 2 by voxelizing 3D data and treating it as multi-directional videos. It supports 3D prompts (point, box, mask) and fuses six directional segmentations to produce coherent 3D masks without training or 2D-3D projection. The approach demonstrates broad generalization across objects, indoor/outdoor scenes, and raw LiDAR, and is positioned as a practical baseline and starting point for promptable 3D segmentation research. The authors provide demonstrations on multiple datasets and release code and an online demo, highlighting potential applications in 3D understanding and multimodal learning.

Abstract

We introduce SAM2Point, a preliminary exploration adapting Segment Anything Model 2 (SAM 2) for zero-shot and promptable 3D segmentation. SAM2Point interprets any 3D data as a series of multi-directional videos, and leverages SAM 2 for 3D-space segmentation, without further training or 2D-3D projection. Our framework supports various prompt types, including 3D points, boxes, and masks, and can generalize across diverse scenarios, such as 3D objects, indoor scenes, outdoor environments, and raw sparse LiDAR. Demonstrations on multiple 3D datasets, e.g., Objaverse, S3DIS, ScanNet, Semantic3D, and KITTI, highlight the robust generalization capabilities of SAM2Point. To our best knowledge, we present the most faithful implementation of SAM in 3D, which may serve as a starting point for future research in promptable 3D segmentation. Online Demo: https://huggingface.co/spaces/ZiyuG/SAM2Point . Code: https://github.com/ZiyuGuo99/SAM2Point .

Figures (7)

• Figure 1: The Segmentation Paradigm of Sam2Point. We introduce a zero-shot and promptable framework for robust 3D segmentation via SAM 2 ravi2024sam. It supports various user-provided 3D prompt, and can generalize to diverse 3D scenarios. The 3D prompt and segmentation results are highlighted in red and green, respectively.
• Figure 2: The Detailed Methodology of Sam2Point. We convert any input 3D data into voxelized representations, and utilize user-provided 3D prompt to divide the 3D space along six directions, effectively simulating six different videos for SAM 2 to perform zero-shot segmentation.
• Figure 3: 3D Object Segmentation with Sam2Point on Objaverse deitke2023objaverse. The 3D prompt and segmentation results are highlighted in red and green, respectively.
• Figure 4: 3D Indoor Scene Segmentation with Sam2Point on S3DIS armeni20163d. The 3D prompt and segmentation results are highlighted in red and green, respectively.
• Figure 5: 3D Indoor Scene Segmentation with Sam2Point on ScanNet dai2017scannet. The 3D prompt and segmentation results are highlighted in red and green, respectively.