Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

KAN See In the Dark

Aoxiang Ning, Minglong Xue, Jinhong He, Chengyun Song

TL;DR

This paper tackles LLIE by addressing nonlinear degradation factors arising from uneven illumination and noise. It introduces KSID, a diffusion-based LLIE framework that embeds a KAN-Block to learn nonlinear mappings with improved interpretability, augmented by a Frequency Domain Perception Module to stabilize the diffusion process and sharpen high-frequency details. The authors provide a Kolmogorov-Arnold theorem grounded rationale for using KANs, detail a two-phase training scheme, and demonstrate strong quantitative and qualitative performance on LLIE benchmarks, including state-of-the-art results on several datasets. The work highlights the potential of interpretable, non-linear architectures in low-level vision tasks and opens avenues for integrating frequency-domain cues into diffusion-based image enhancement.

Abstract

Existing low-light image enhancement methods are difficult to fit the complex nonlinear relationship between normal and low-light images due to uneven illumination and noise effects. The recently proposed Kolmogorov-Arnold networks (KANs) feature spline-based convolutional layers and learnable activation functions, which can effectively capture nonlinear dependencies. In this paper, we design a KAN-Block based on KANs and innovatively apply it to low-light image enhancement. This method effectively alleviates the limitations of current methods constrained by linear network structures and lack of interpretability, further demonstrating the potential of KANs in low-level vision tasks. Given the poor perception of current low-light image enhancement methods and the stochastic nature of the inverse diffusion process, we further introduce frequency-domain perception for visually oriented enhancement. Extensive experiments demonstrate the competitive performance of our method on benchmark datasets. The code will be available at: https://github.com/AXNing/KSID}{https://github.com/AXNing/KSID.

KAN See In the Dark

TL;DR

This paper tackles LLIE by addressing nonlinear degradation factors arising from uneven illumination and noise. It introduces KSID, a diffusion-based LLIE framework that embeds a KAN-Block to learn nonlinear mappings with improved interpretability, augmented by a Frequency Domain Perception Module to stabilize the diffusion process and sharpen high-frequency details. The authors provide a Kolmogorov-Arnold theorem grounded rationale for using KANs, detail a two-phase training scheme, and demonstrate strong quantitative and qualitative performance on LLIE benchmarks, including state-of-the-art results on several datasets. The work highlights the potential of interpretable, non-linear architectures in low-level vision tasks and opens avenues for integrating frequency-domain cues into diffusion-based image enhancement.

Abstract

Existing low-light image enhancement methods are difficult to fit the complex nonlinear relationship between normal and low-light images due to uneven illumination and noise effects. The recently proposed Kolmogorov-Arnold networks (KANs) feature spline-based convolutional layers and learnable activation functions, which can effectively capture nonlinear dependencies. In this paper, we design a KAN-Block based on KANs and innovatively apply it to low-light image enhancement. This method effectively alleviates the limitations of current methods constrained by linear network structures and lack of interpretability, further demonstrating the potential of KANs in low-level vision tasks. Given the poor perception of current low-light image enhancement methods and the stochastic nature of the inverse diffusion process, we further introduce frequency-domain perception for visually oriented enhancement. Extensive experiments demonstrate the competitive performance of our method on benchmark datasets. The code will be available at: https://github.com/AXNing/KSID}{https://github.com/AXNing/KSID.
Paper Structure (15 sections, 9 equations, 4 figures, 2 tables)

This paper contains 15 sections, 9 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Kolmogorov-Arnold theorem Preliminaries
  4. Overall Network Architecture
  5. KAN-Block
  6. Frequency Domain Perception Module
  7. Experiments
  8. Experimental Settings
  9. Datasets and Metrics
  10. Implementation Details
  11. Comparisons With State-of-the-Art Methods
  12. Quantitative Comparisons
  13. Qualitative Comparisons
  14. Ablation Study
  15. Conclusion

Figures (4)

  • Figure 1: Our method effectively learns the nonlinear degradation factors in the low-light domain, especially in darker scenes, and our recovery significantly improves compared to the GSAD.
  • Figure 2: (a) Illustration of the training workflow of the proposed method. (b) Detailed of the KAN-Block. (c) Structure of the Kolmogorov-Arnold Networks. $X$ is the input feature
  • Figure 3: Visual comparisons of the enhanced results by different methods on LOLv2.
  • Figure 4: A visual comparison of results with and without the Frequency Domain Perception Module.