Language-Assisted Super-Resolution from Real-World Low-Resolution Patches

2026-06-30

AI Summary

This work addresses the limited generalization of super-resolution models caused by the scarcity of real-world paired high- and low-resolution images. To overcome this, we propose an unpaired super-resolution approach that leverages a single high-quality image containing naturally occurring high- and low-resolution regions due to depth-of-field variations, from which authentic low-resolution patches are extracted. We reformulate the task as a cross-modal alignment problem in a vision–language semantic space, a novel perspective not previously explored. High-fidelity and semantically consistent reconstructions are achieved through content and quality losses guided by a vision–language model. Experiments demonstrate that our method significantly outperforms existing approaches trained on synthetically degraded data when applied to real low-resolution images.

Abstract

Single image super-resolution aims to reconstruct high-resolution (HR) images from low-resolution (LR) inputs. Training SR models typically requires paired HR-LR data, which is difficult to obtain in reality. As a result, most methods synthesize LR images by artificially degrading HR images with handcrafted kernels or camera ISP adjustments. However, these synthetic degradations fail to capture the complexity of real LR images, leading to poor generalization in practice. To address this, we observe that even within a single high-quality image, regions at different depths exhibit varying resolutions, where distant regions act as LR patches and closer ones as HR patches. This allows the extraction of real, degradation-induced LR patches from real images. Since these LR patches lack paired HR counterparts, we propose LA-SR (Language Assistant for SR), a novel framework for unpaired SR. The key idea of LA-SR is to redefine unpaired SR in the language space, using vision-language models to bridge the LR-HR gap. LA-SR projects images into a semantically rich space representing both content and quality, and applies two language-guided losses: linguistic content loss to preserve semantic fidelity, and linguistic quality loss to enhance perceptual realism. With this alignment, LA-SR effectively super-resolves real LR inputs, producing realistic outputs that overcome the limitations of synthetic-data-trained methods.

Problem

• super-resolution
• real-world degradation
• unpaired image restoration
• low-resolution patches
• image synthesis

Innovation

• language-assisted super-resolution
• unpaired image super-resolution
• vision-language models
• real-world degradation
• semantic alignment