Scholarlink

<p>This study introduces a novel 3D segmentation framework leveraging 2D Gaussian Splatting for precise geometric reconstruction. By integrating the Segment Anything Model (SAM) with multi-view consistency tracking and text prompting, we generate refined 2D masks to guide 3D segmentation. Our approach achieves fine-grained segmentation, improving training efficiency by over 44% and enabling rapid acquisition of high-quality meshes. Experimental results on diverse datasets demonstrate robust segmentation accuracy, outperforming existing methods in both segmentation quality and computational efficiency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced 3D segmentation via 2D Gaussian Splatting and multi-view consistency

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Zhiyan Tai,
  • Wenhao Huang,
  • Zhuhong Shao,
  • Shengze Yu,
  • Ming Liu

摘要

This study introduces a novel 3D segmentation framework leveraging 2D Gaussian Splatting for precise geometric reconstruction. By integrating the Segment Anything Model (SAM) with multi-view consistency tracking and text prompting, we generate refined 2D masks to guide 3D segmentation. Our approach achieves fine-grained segmentation, improving training efficiency by over 44% and enabling rapid acquisition of high-quality meshes. Experimental results on diverse datasets demonstrate robust segmentation accuracy, outperforming existing methods in both segmentation quality and computational efficiency.