Scholarlink

<p>In dental clinical practice, the use of digital three-dimensional surface models is increasingly prevalent. However, high-resolution dental meshes pose significant storage and computational challenges. This paper introduces a dual-domain neural compression (DDNC) framework for dental mesh compression, leveraging both the geometric and neural domains to achieve high compression ratios while preserving critical dental features. The proposed method first generates a coarse mesh representation, then predicts fine-grained displacements through a neural network, and finally performs multi-level optimization in a coarse-to-fine manner to achieve high-fidelity and visually coherent surface refinement. Experimental results demonstrate superior performance in detail preservation and compression efficiency compared to traditional and learning-based methods, achieving a 99% reduction in file size with minimal loss of geometric detail. Code is available at: <a href="https://github.com/CCShermit1/DDNC">https://github.com/CCShermit1/DDNC</a></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

DDNC: dual-domain neural compression for high-fidelity dental mesh

  • 期刊论文

  • 发表日期：

    2026年2月21日

查看全文
  • Yuteng Liu,
  • Yutong Hu,
  • Haisheng Li,
  • Li Chen

摘要

In dental clinical practice, the use of digital three-dimensional surface models is increasingly prevalent. However, high-resolution dental meshes pose significant storage and computational challenges. This paper introduces a dual-domain neural compression (DDNC) framework for dental mesh compression, leveraging both the geometric and neural domains to achieve high compression ratios while preserving critical dental features. The proposed method first generates a coarse mesh representation, then predicts fine-grained displacements through a neural network, and finally performs multi-level optimization in a coarse-to-fine manner to achieve high-fidelity and visually coherent surface refinement. Experimental results demonstrate superior performance in detail preservation and compression efficiency compared to traditional and learning-based methods, achieving a 99% reduction in file size with minimal loss of geometric detail. Code is available at: https://github.com/CCShermit1/DDNC