Scholarlink

<p>The reconstruction of textured 3D human models from a single image is a fundamental task with wide-ranging applications in fields such as virtual reality, gaming, and digital human rendering. Current approaches, however, are often time-consuming and struggle to accurately capture detailed texture information, limiting their practicality in real-time applications. To address this, we propose a novel method, <i>Ultraman</i>, which significantly accelerates the reconstruction process while enhancing the accuracy and preserving high-quality texture details. Our method is built on three core modules: (1) a geometric reconstruction framework that extracts accurate 3D human shapes from a single image; (2) a texture generation approach that produces multi-view consistent images based on the input image; and (3) a novel texture mapping technique that optimizes texture details and ensures color consistency throughout the reconstruction process. Extensive experiments on standard datasets demonstrate that Ultraman outperforms state-of-the-art methods in both rendering quality and speed, establishing a new benchmark for fast, high-fidelity 3D human reconstruction. Our contributions include the introduction of a new framework for efficient and accurate texture generation, as well as a robust evaluation showing significant improvements over existing techniques. Code is available for research purposes at <a href="https://air-discover.github.io/Ultraman/">https://air-discover.github.io/Ultraman/</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

Ultraman: ultra-fast and high-resolution texture generation for 3D human reconstruction from a single image

  • 期刊论文

  • 发表日期：

    2026年1月21日

查看全文
  • Mingjin Chen,
  • Junhao Chen,
  • Huan-ang Gao,
  • Xiaoxue Chen,
  • Zhaoxin Fan,
  • Hao Zhao

摘要

The reconstruction of textured 3D human models from a single image is a fundamental task with wide-ranging applications in fields such as virtual reality, gaming, and digital human rendering. Current approaches, however, are often time-consuming and struggle to accurately capture detailed texture information, limiting their practicality in real-time applications. To address this, we propose a novel method, Ultraman, which significantly accelerates the reconstruction process while enhancing the accuracy and preserving high-quality texture details. Our method is built on three core modules: (1) a geometric reconstruction framework that extracts accurate 3D human shapes from a single image; (2) a texture generation approach that produces multi-view consistent images based on the input image; and (3) a novel texture mapping technique that optimizes texture details and ensures color consistency throughout the reconstruction process. Extensive experiments on standard datasets demonstrate that Ultraman outperforms state-of-the-art methods in both rendering quality and speed, establishing a new benchmark for fast, high-fidelity 3D human reconstruction. Our contributions include the introduction of a new framework for efficient and accurate texture generation, as well as a robust evaluation showing significant improvements over existing techniques. Code is available for research purposes at https://air-discover.github.io/Ultraman/.