Scholarlink

<p>This paper presents PointGF, a geometry-enhanced deep learning framework for point cloud classification and semantic segmentation. The proposed framework incorporates a geometry-aware affine transformation module that leverages local structural statistics to improve feature alignment in point cloud representations. A curvature-based adaptive sampling strategy is further integrated into the learning pipeline to better preserve geometrically informative regions during feature extraction. The proposed modules are designed as plug-in components that can be integrated into existing point-based networks to enhance geometric representation learning. Experimental evaluations on ModelNet40, ScanObjectNN, and S3DIS show that PointGF achieves competitive performance in both classification and segmentation tasks, demonstrating an effective trade-off between improved accuracy and additional computational overhead.</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 point cloud processing through geometric affine transformations and curvature-based sampling

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Yuxin Zhang,
  • Tao Zuo,
  • Yanlong Lu,
  • Zhaowei Wu

摘要

This paper presents PointGF, a geometry-enhanced deep learning framework for point cloud classification and semantic segmentation. The proposed framework incorporates a geometry-aware affine transformation module that leverages local structural statistics to improve feature alignment in point cloud representations. A curvature-based adaptive sampling strategy is further integrated into the learning pipeline to better preserve geometrically informative regions during feature extraction. The proposed modules are designed as plug-in components that can be integrated into existing point-based networks to enhance geometric representation learning. Experimental evaluations on ModelNet40, ScanObjectNN, and S3DIS show that PointGF achieves competitive performance in both classification and segmentation tasks, demonstrating an effective trade-off between improved accuracy and additional computational overhead.