Linear Recurrent Unit with Semantic Modulation for Image Super-Resolution

2026-06-18

AI Summary

This work addresses the performance limitations of Linear Recurrent Units (LRUs) in two-dimensional vision tasks, which stem from static parameterization and unidirectional scanning. To overcome these bottlenecks, the authors propose a novel LRU architecture integrated with a Semantic Modulation Unit (SMU). This approach introduces semantic modulation into LRUs for the first time, enabling dynamic feature modulation, spatial region partitioning, and prototype-based feature enhancement learned from data. The proposed model achieves significantly improved reconstruction quality in single-image super-resolution while maintaining computational complexity comparable to existing methods. Extensive experiments demonstrate that the model consistently outperforms state-of-the-art approaches both quantitatively and qualitatively across multiple benchmarks, effectively striking a balance between performance and efficiency.

Abstract

Linear recurrent unit (LRU), designed with a principled formulation for stable linear recurrence, has demonstrated promising accuracy and robustness on long-range dependency tasks. However, its static parameterization and single-scan method limits its applicability to 2D vision tasks. In this study, we propose a LRU-based restoration network with a semantic modulating unit (SMU) to achieve a harmonious balance between performance and efficiency in single-image super-resolution. The SMU plays three key roles: LRU modulation, spatial categorization, and feature enhancement through learned prototype. Extensive experiments demonstrate that our method quantitatively and qualitatively surpasses recent state-of-the-art methods. Notably, our approach achieves superior performance with computational complexity on par with existing methods. The source code and models are available at https://github.com/MingyuChoi-run/LSM

Problem

• Linear Recurrent Unit
• Image Super-Resolution
• Static Parameterization
• Single-Scan Method
• 2D Vision Tasks

Innovation

• Linear Recurrent Unit
• Semantic Modulation
• Image Super-Resolution
• Feature Enhancement
• Spatial Categorization