Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DeRainMamba: A Frequency-Aware State Space Model with Detail Enhancement for Image Deraining

Zhiliang Zhu, Tao Zeng, Tao Yang, Guoliang Luo, Jiyong Zeng

TL;DR

DeRainMamba tackles image deraining by addressing the need for both global context and fine local details. It introduces a Frequency-Aware State Space Model (FASSM) that fuses a Vision State Space Module (VSSM) with a Residual Fourier Module (RFM) to jointly model spatial context and frequency-domain cues, using the fusion expression $F_{out} = \text{VSSM}(\text{LN}(F_{in})) + \text{RFM}(F_{in}) + s \cdot F_{in}$. It also introduces Multi-direction Perception Convolution (MDPConv) to capture gradient-based details with a single efficient kernel, described by $F_{out} = \text{MDPConv}(F_{in}) = F_{in} * K_{eq}$ (with $K_{eq} = \sum_{i=1}^5 K_i$). The training optimizes $\mathcal{L}_{total}= \lambda_{1} \mathcal{L}_{1} + \lambda_{2} \mathcal{L}_{Freq}$ with $\lambda_{1}=1$ and $\lambda_{2}=0.1$, and experiments on four benchmarks show improvements with a lightweight model (~27.8M parameters), validating the efficacy of combining frequency-aware global modeling with local gradient-based detail enhancement for deraining.

Abstract

Image deraining is crucial for improving visual quality and supporting reliable downstream vision tasks. Although Mamba-based models provide efficient sequence modeling, their limited ability to capture fine-grained details and lack of frequency-domain awareness restrict further improvements. To address these issues, we propose DeRainMamba, which integrates a Frequency-Aware State-Space Module (FASSM) and Multi-Directional Perception Convolution (MDPConv). FASSM leverages Fourier transform to distinguish rain streaks from high-frequency image details, balancing rain removal and detail preservation. MDPConv further restores local structures by capturing anisotropic gradient features and efficiently fusing multiple convolution branches. Extensive experiments on four public benchmarks demonstrate that DeRainMamba consistently outperforms state-of-the-art methods in PSNR and SSIM, while requiring fewer parameters and lower computational costs. These results validate the effectiveness of combining frequency-domain modeling and spatial detail enhancement within a state-space framework for single image deraining.

DeRainMamba: A Frequency-Aware State Space Model with Detail Enhancement for Image Deraining

TL;DR

DeRainMamba tackles image deraining by addressing the need for both global context and fine local details. It introduces a Frequency-Aware State Space Model (FASSM) that fuses a Vision State Space Module (VSSM) with a Residual Fourier Module (RFM) to jointly model spatial context and frequency-domain cues, using the fusion expression . It also introduces Multi-direction Perception Convolution (MDPConv) to capture gradient-based details with a single efficient kernel, described by (with ). The training optimizes with and , and experiments on four benchmarks show improvements with a lightweight model (~27.8M parameters), validating the efficacy of combining frequency-aware global modeling with local gradient-based detail enhancement for deraining.

Abstract

Image deraining is crucial for improving visual quality and supporting reliable downstream vision tasks. Although Mamba-based models provide efficient sequence modeling, their limited ability to capture fine-grained details and lack of frequency-domain awareness restrict further improvements. To address these issues, we propose DeRainMamba, which integrates a Frequency-Aware State-Space Module (FASSM) and Multi-Directional Perception Convolution (MDPConv). FASSM leverages Fourier transform to distinguish rain streaks from high-frequency image details, balancing rain removal and detail preservation. MDPConv further restores local structures by capturing anisotropic gradient features and efficiently fusing multiple convolution branches. Extensive experiments on four public benchmarks demonstrate that DeRainMamba consistently outperforms state-of-the-art methods in PSNR and SSIM, while requiring fewer parameters and lower computational costs. These results validate the effectiveness of combining frequency-domain modeling and spatial detail enhancement within a state-space framework for single image deraining.

Paper Structure

This paper contains 11 sections, 5 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Frequency-Aware State Space Model
  4. Multi-direction Perception Convolution
  5. Loss Function
  6. Experiments
  7. Experimental Settings
  8. Quantitative and Qualitative Results
  9. Comparison of model complexity
  10. Ablation Study
  11. Conclusion

Figures (3)

  • Figure 1: Visualization of phase differences between rain-free and rainy images.
  • Figure 2: Overall architecture of the proposed DeRainMamba. The network consists of multiple Frequency Gradient Mamba (FGM) blocks, each comprising a Frequency-Aware State Space Module (FASSM) for global modeling and an MDPConv branch for gradient-based detail enhancement. The two branches are fused to jointly suppress rain streaks and restore image details.
  • Figure 3: Visual comparison on the Rain200H dataset.