MBMamba: When Memory Buffer Meets Mamba for Structure-Aware Image Deblurring
Hu Gao, Xiaoning Lei, Xichen Xu, Depeng Dang, Lizhuang Ma

TL;DR
MBMamba enhances the Mamba architecture for image deblurring by introducing a memory buffer and Ising-inspired regularization, effectively preserving image structure and outperforming existing methods without increasing computational complexity.
Contribution
It proposes a structure-aware deblurring network that maintains the original Mamba architecture while adding a memory buffer and a novel regularization loss for better spatial information modeling.
Findings
Outperforms state-of-the-art on benchmark datasets.
Effectively preserves image structure during deblurring.
Maintains real-time performance without added complexity.
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…
Peer Reviews
Decision·ICLR 2026 Conference Withdrawn Submission
1. Elegant structural innovation. The proposed memory buffer mechanism cleverly leverages temporal-like dependency modeling to retain local context within the Mamba framework, without modifying the core architecture. This is a nontrivial conceptual advancement, positioning MBMamba as a “plug-in” enhancement rather than a structural overhaul. 2. Ising-inspired loss bridging physics and perception. The introduction of the Ising regularizer is an intriguing idea that connects statistical physics wi
1. Lack of theoretical or mathematical grounding of the memory buffer mechanism The design is described algorithmically but not theoretically. How does the memory depth K influence the state propagation in Mamba’s continuous SSM formulation? Is the memory fusion equivalent to modifying the recurrent transition kernel? Does this break the linearity and stability guarantees of Mamba’s state evolution? A deeper analysis (e.g., spectrum or stability study) would elevate the contribution from empiric
The paper is well written and easy to follow. The motivation for introducing a memory buffer is clearly presented, and its effectiveness is well supported by extensive experimental results.
1. Task specificity The proposed method is not specifically tailored to the image deblurring task. To demonstrate its generalization capability, the model should also be evaluated on other common image restoration tasks, such as image denoising and super-resolution. 2. Memory mechanism design The memory management strategy appears overly simplistic. The FIFO-based memory update may lead to the accumulation of redundant features in the memory buffer, which could eventually result in performance
The MemVSSM module is designed to enhance local context preservation via feature partitioning and temporal buffering. The physically inspired Ising Loss is incorporated into the deblurring loss function, enforcing spatial coherence through energy minimization. Notably, this module extends the spatial-awareness capability of the Mamba backbone without altering its main architecture, achieving both efficiency and accuracy. The proposed method provides a new perspective on memory-based state-space
The title and abstract emphasize “Structure-Aware”, yet the manuscript does not present any structural-awareness visualizations or heatmaps (No direct evidence found in the manuscript). The paper does not discuss how the depth (K) of MemVSSM or the number of channel partitions (N) affects performance. Similarly, no comparison is provided against a non-memory Mamba baseline trained in a single step. Sections 3.2 and 3.3 describe MemVSSM and the Ising Loss separately but do not explain their tra
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Image Processing Techniques · Digital Media Forensic Detection · Image and Signal Denoising Methods
