Search papers, labs, and topics across Lattice.
This paper introduces GraftSR, a novel generative framework for real-world image super-resolution that addresses the issue of texture hallucination by utilizing identical-instance reference images for texture restoration. The method employs a dual-mask reference guidance mechanism to effectively decouple texture extraction from spatial alignment challenges, allowing for precise application of authentic textures. Experimental results on the TexRefSR-Eval benchmark show that GraftSR achieves a 20.2% reduction in LPIPS compared to leading baselines, establishing a new state-of-the-art in reference-faithful image restoration.
GraftSR achieves a groundbreaking 20.2% reduction in perceptual error, setting a new benchmark for texture fidelity in image super-resolution.
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.
