MBMamba: When Memory Buffer Meets Mamba for Structure-Aware Image Deblurring

2025-08-17

AI Summary

MBMamba addresses two key limitations of the Mamba architecture in image deblurring: local pixel forgetting and channel redundancy caused by sequential row-wise flattening. To this end, it proposes a structural-aware, lightweight enhancement that preserves the original Mamba backbone unchanged. Specifically, it introduces a learnable memory buffer to explicitly retain and fuse multi-scale spatial features, and designs an Ising model-inspired structural regularization loss to model local pixel correlations and enforce structural coherence. The method significantly enhances 2D spatial modeling capability while maintaining real-time inference speed. Extensive experiments demonstrate state-of-the-art performance on major benchmarks including GoPro and HIDE, with particularly notable improvements in edge sharpness and texture recovery.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Structured Learning
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

The Mamba architecture has emerged as a promising alternative to CNNs and Transformers for image deblurring. However, its flatten-and-scan strategy often results in local pixel forgetting and channel redundancy, limiting its ability to effectively aggregate 2D spatial information. Although existing methods mitigate this by modifying the scan strategy or incorporating local feature modules, it increase computational complexity and hinder real-time performance. In this paper, we propose a structure-aware image deblurring network without changing the original Mamba architecture. Specifically, we design a memory buffer mechanism to preserve historical information for later fusion, enabling reliable modeling of relevance between adjacent features. Additionally, we introduce an Ising-inspired regularization loss that simulates the energy minimization of the physical system's "mutual attraction" between pixels, helping to maintain image structure and coherence. Building on this, we develop MBMamba. Experimental results show that our method outperforms state-of-the-art approaches on widely used benchmarks.

Problem

• Address local pixel forgetting in Mamba for image deblurring
• Reduce channel redundancy in Mamba's spatial information aggregation
• Maintain real-time performance while improving deblurring accuracy

Innovation

• Memory buffer mechanism preserves historical information
• Ising-inspired regularization loss maintains image structure
• MBMamba combines memory buffer with Mamba architecture