[论文] SUPRA-TEX: 3D Generation with Super-Aligned Texture

论文概要

研究领域: CV 作者: Chen Yang, Jiemin Fang, Fanpeng Meng, Zihao Huang, Jichen Pang, Guanjun Wu, Zhikuan Bao, Qi Tian 发布时间: 2026-10-15 arXiv: 2610.22352

中文摘要

单图 3D 生成旨在为给定 3D 网格赋予可重光照材质，同时忠实保留参考图像的视觉细节。现有方法在融合不一致的多视角预测时引入伪影，或通过隐式图像特征条件难以保留细节。我们引入 SUPRA-Tex，建立在解耦可见材质估计、表面对齐和观测引导 3D 补全的原则上。首先训练基于 MLLM 的图像编辑模型，在参考视图中预测光照分离的 PBR 贴图，然后投影到可见表面并编码为与生成目标共享潜空间的观测潜变量。交错的流 Transformer 将这些潜变量作为固定只读锚点，交替进行未观测区域和全表面更新，生成完整材质潜变量。仅在公共数据集上训练，SUPRA-Tex 在几何条件纹理和端到端资产生成中均达到 SOTA，匹配甚至超过领先商业系统。项目页面：https://supra-3d.github.io/supratex/

原文摘要

Single image 3D generation aims to dress a given 3D mesh with relightable materials while faithfully preserving the visual details of a reference image. Existing methods either introduce artifacts when fusing inconsistent multi-view predictions or struggle to preserve fine details through implicit image-feature conditioning. To resolve these challenges, we introduce SUPRA-Tex, built on the principle of decoupling visible-material estimation, surface alignment, and observation-guided 3D completion. This division of labor allows the 3D generator, trained on limited 3D data, to focus its capacity on preserving visible evidence and completing unobserved surfaces. Specifically, we first train an MLLM-based image editing model to predict lighting-separated PBR maps in the reference view. We then...

#论文 #arXiv #CV #小凯