MixNet: Efficient Global Modeling for Ultra-High-Definition Image Restoration

TL;DR

MixNet addresses the challenge of efficient UHD image restoration by introducing a permutation-based Global Feature Modulation Layer (GFML) that enables long-range modeling without heavy computation. It fuses GFML with Local Feature Modulation Layer (LFML) and a Feed-forward Layer (FFL) inside Feature Mixing Blocks to balance global and local information while maintaining compact representations for full-resolution UHD inference. Across four UHD restoration tasks—low-light enhancement, underwater enhancement, deblurring, and demoiréing—the method achieves state-of-the-art or competitive results with lower inference costs than transformer-based approaches. The work demonstrates practical potential for real-time UHD restoration on consumer hardware and provides code for reproducibility.

Abstract

Recent advancements in image restoration methods employing global modeling have shown promising results. However, these approaches often incur substantial memory requirements, particularly when processing ultra-high-definition (UHD) images. In this paper, we propose a novel image restoration method called MixNet, which introduces an alternative approach to global modeling approaches and is more effective for UHD image restoration. To capture the longrange dependency of features without introducing excessive computational complexity, we present the Global Feature Modulation Layer (GFML). GFML associates features from different views by permuting the feature maps, enabling efficient modeling of long-range dependency. In addition, we also design the Local Feature Modulation Layer (LFML) and Feed-forward Layer (FFL) to capture local features and transform features into a compact representation. This way, our MixNetachieves effective restoration with low inference time overhead and computational complexity. We conduct extensive experiments on four UHD image restoration tasks, including low-light image enhancement, underwater image enhancement, image deblurring and image demoireing, and the comprehensive results demonstrate that our proposed method surpasses the performance of current state-of-the-art methods. The code will be available at \url{https://github.com/5chen/MixNet}.

Figures (8)

• Figure 1: Our MixNet significantly outperforms state-of-the-art UHD image restoration methods, including UHD, UHDFour, UHDformer and ESDNet, on four UHD image restoration benchmarks.
• Figure 2: The overall architecture of the proposed MixiNet. MixNet first transforms the input degraded image into the feature space using a DownSampler, performs feature extraction using a series of Feature Mixing Blocks (FMBs), and then reconstructs these extracted features using an UpSampler. The FMB mainly contains a Global Feature Modulation Layer (GFML), a Local Feature Modulation Layer (LFML), and a Feed-forward Layer (FFL).
• Figure 3: Visualization of dimensional transformation operations. MixNet can capture long-range dependency of features with few parameters by employing simple dimension transformation operations.
• Figure 4: Visual quality comparisons with state-of-the-art methods on UHD-LOL4K dataset. The last row shows the color histogram of the image. Our method has the closest color to GT image.
• Figure 5: Visual quality comparisons with state-of-the-art methods on UHD-UW dataset. Please zoom in for details.