OmniFabric: Coherent UV Space Texture Synthesis for 3D Garment Reconstruction from a Single Image

Forum topic · 小凯 · 2026-09-28

Summary

OmniFabric is a new computer vision method for generating production-ready 3D garment assets from a single reference image, addressing the long-standing bottleneck of texture synthesis. While generative models have advanced 3D geometry reconstruction, existing texture pipelines often bake environmental illumination and shadows into texture maps or fail to maintain global structural coherence, making assets unusable for physical simulation and relighting. OmniFabric synthesizes globally coherent texture maps directly in the 2D sewing pattern space. The pipeline first uses an estimated 3D mesh and generative priors from vision-language models (VLMs) to create a complete but coarse texture initialization on the unfolded sewing pattern. A specialized diffusion Transformer—trained with an automated synthetic data engine and conditioned on 3D positional features—then refines the initialization directly in the standard UV domain, effectively removing distortion and baked-in artifacts to extract clean, canonical texture maps that preserve the original garment design. Extensive experiments show OmniFabric significantly outperforms state-of-the-art baselines, producing photorealistic 3D garments with high-quality textures. The paper (arXiv:2609.30234) is by Ding-Jiun Huang, Yuanhao Wang, Cheng Zhang, Hugo Bertiche, Alexandru-Eugen Ichim, Thabo Beeler, and Fernando De la Torre.

Paper Overview

Research area: Computer Vision Authors: Ding-Jiun Huang, Yuanhao Wang, Cheng Zhang, Hugo Bertiche, Alexandru-Eugen Ichim, Thabo Beeler, Fernando De la Torre Published: 2026-09-24 arXiv: 2609.30234

Summary

Automated generation of production-ready 3D garment assets from a single image is a central challenge in digital content creation. While recent generative models have significantly advanced 3D geometry reconstruction, synthesizing high-quality textures remains a bottleneck. Existing methods often bake environmental illumination and shadows directly into the texture map, or they fail to maintain global structural coherence, making the resulting assets unusable for physical simulation and relighting.

OmniFabric introduces a novel approach that synthesizes globally coherent texture maps directly within the 2D sewing pattern space. Given a single reference image, the pipeline:

1. Utilizes an estimated 3D mesh and generative priors of powerful Vision-Language Models (VLM) to establish a complete but coarse texture initialization on the unfolded sewing pattern. 2. Refines this initialization directly in the standard UV domain using a specialized diffusion Transformer, trained with an automated synthetic data engine and conditioned on 3D positional features.

This process effectively eliminates distortion and baked-in artifacts, extracting clean and canonical texture maps that preserve the original garment design. Extensive experiments demonstrate that OmniFabric significantly outperforms state-of-the-art baselines, producing photorealistic 3D garments with high-quality textures suitable for physical simulation and relighting.

--- *Auto-collected on 2026-09-28.*

OmniFabric: Coherent UV Space Texture Synthesis for 3D Garment Reconstruction from a Single Image

Summary

Paper Overview

Summary

Tags