Scholarlink

<p>With the rise of digital technologies, three-dimensional (3D) garments have become indispensable in virtual fashion showcases, fitting rooms, and in-game avatars. This paper presents DeepSketch2Wear, a method that democratizes 3D garment creation by converting users’ freehand sketches and text descriptions into detailed 3D garment models with textures. Leveraging two conditional diffusion models, we generate high-quality Unsigned Distance Fields in a coarse-to-fine manner and a diffusion model for texture coordinate textures. Sketch enhancement techniques are incorporated to improve robustness against varying drawing skills. Extensive experiments demonstrate our method’s effectiveness in terms of quality and fidelity across synthetic and real-world benchmarks, with users expressing greater satisfaction compared to existing methods. Here we show that our approach achieves state-of-the-art performance, with a Chamfer Distance of 0.011 and a Kernel Inception Distance of 3.097 on the KO3DG dataset, significantly outperforming existing techniques. This work paves the way for richer user experiences in digital environments and human-computer interaction. The source code is available at <a href="https://github.com/cansinyu/DeepSketch2Wear">https://github.com/cansinyu/DeepSketch2Wear</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

DeepSketch2Wear: democratizing 3D garment creation via freehand sketches and text

  • 期刊论文

  • 发表日期：

    2026年1月19日

查看全文
  • Xinyu Chen,
  • Jianqi Zhang,
  • Chaotao Ding,
  • Runlong Cao,
  • Lanyun Zhu,
  • Wenjun Hu,
  • Ying Zang,
  • Tianrun Chen

摘要

With the rise of digital technologies, three-dimensional (3D) garments have become indispensable in virtual fashion showcases, fitting rooms, and in-game avatars. This paper presents DeepSketch2Wear, a method that democratizes 3D garment creation by converting users’ freehand sketches and text descriptions into detailed 3D garment models with textures. Leveraging two conditional diffusion models, we generate high-quality Unsigned Distance Fields in a coarse-to-fine manner and a diffusion model for texture coordinate textures. Sketch enhancement techniques are incorporated to improve robustness against varying drawing skills. Extensive experiments demonstrate our method’s effectiveness in terms of quality and fidelity across synthetic and real-world benchmarks, with users expressing greater satisfaction compared to existing methods. Here we show that our approach achieves state-of-the-art performance, with a Chamfer Distance of 0.011 and a Kernel Inception Distance of 3.097 on the KO3DG dataset, significantly outperforming existing techniques. This work paves the way for richer user experiences in digital environments and human-computer interaction. The source code is available at https://github.com/cansinyu/DeepSketch2Wear.