发表机构
Taobao & Tmall Group of Alibaba; University of Chinese Academy of Sciences（阿里巴巴淘宝天猫集团; 中国科学院大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对扩散图像超分辨率的纹理幻觉问题，提出GraftSR框架，采用双掩码参考引导机制，构建TexRefSR-141K数据集，在新基准上实现SOTA，LPIPS降低20.2%。
AI 中文摘要
基于扩散模型的真实世界图像超分辨率（SR）方法能实现出色的感知质量，但固有存在严重的纹理幻觉问题。为克服这一局限，我们提出GraftSR，一种由纹理参考引导的生成式SR框架，利用相同实例的参考图像来锚定真实纹理的恢复。然而，低质量输入与其参考图像之间存在严重的空间错位，带来了重大挑战，常导致模糊的迁移目标和背景特征泄漏。为解决这些问题，GraftSR采用了一种新颖的双掩码参考引导机制，系统地解耦跨视角纹理注入过程。通过明确区分从参考图像中提取哪些真实纹理，并精确地将其应用于目标图像的何处，GraftSR实现了稳健的纹理迁移，无需依赖脆弱的空间对齐。此外，为弥合合适训练数据的关键缺口，我们构建了TexRefSR-141K，这是首个提供带有互补空间掩码的高质量参考元组的大规模数据集。在我们新建立的基准TexRefSR-Eval上进行的大量实验表明，GraftSR达到了新的最先进水平。值得注意的是，它相比表现最佳的基线方法降低了20.2%的LPIPS值，实现了更出色的参考忠实恢复效果。
英文摘要
Diffusion-based real-world image super-resolution (SR) achieves impressive perceptual quality but inherently suffers from severe texture hallucination. To overcome this limitation, we propose GraftSR, a texture-reference-guided generative SR framework that leverages reference images of the identical instance to anchor the restoration of authentic textures. However, severe spatial misalignment between low-quality inputs and their references poses significant challenges, often leading to ambiguous transfer targets and background feature leakage. To address these issues, GraftSR employs a novel dual-mask reference guidance mechanism that systematically decouples the cross-view texture injection process. By explicitly isolating what authentic textures to extract from the reference and precisely localizing where to apply them within the target, GraftSR achieves robust texture transfer without relying on brittle spatial alignment. Furthermore, to bridge the critical gap in appropriate training data, we construct TexRefSR-141K, the first large-scale dataset providing high-quality reference tuples equipped with complementary spatial masks. Extensive experiments on our newly established benchmark, TexRefSR-Eval, demonstrate that GraftSR sets a new state-of-the-art. Notably, it reduces LPIPS by 20.2\% over top-performing baselines, achieving superior reference-faithful restoration.
Comments15 pages, 12 figures
