Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unpaired Image Dehazing via Kolmogorov-Arnold Transformation of Latent Features

Le-Anh Tran

TL;DR

This paper tackles unpaired image dehazing by leveraging Kolmogorov-Arnold Networks (KANs) to transform latent features, circumventing the need for paired hazy-clean data. It introduces a Dual-GR-KAN Transformer within a ViT-inspired generator and employs adversarial training plus patch-wise contrastive learning to enforce semantic and structural fidelity. The approach achieves state-of-the-art performance on RESIDE-derived benchmarks with notably lower computational cost, and its variants offer favorable speed-accuracy trade-offs. The work demonstrates the potential of KANs for ill-posed image restoration tasks and provides code for reproducibility.

Abstract

This paper proposes an innovative framework for Unsupervised Image Dehazing via Kolmogorov-Arnold Transformation, termed UID-KAT. Image dehazing is recognized as a challenging and ill-posed vision task that requires complex transformations and interpretations in the feature space. Recent advancements have introduced Kolmogorov-Arnold Networks (KANs), inspired by the Kolmogorov-Arnold representation theorem, as promising alternatives to Multi-Layer Perceptrons (MLPs) since KANs can leverage their polynomial foundation to more efficiently approximate complex functions while requiring fewer layers than MLPs. Motivated by this potential, this paper explores the use of KANs combined with adversarial training and contrastive learning to model the intricate relationship between hazy and clear images. Adversarial training is employed due to its capacity in producing high-fidelity images, and contrastive learning promotes the model's emphasis on significant features while suppressing the influence of irrelevant information. The proposed UID-KAT framework is trained in an unsupervised setting to take advantage of the abundance of real-world data and address the challenge of preparing paired hazy/clean images. Experimental results show that UID-KAT achieves state-of-the-art dehazing performance across multiple datasets and scenarios, outperforming existing unpaired methods while reducing model complexity. The source code for this work is publicly available at https://github.com/tranleanh/uid-kat.

Unpaired Image Dehazing via Kolmogorov-Arnold Transformation of Latent Features

TL;DR

This paper tackles unpaired image dehazing by leveraging Kolmogorov-Arnold Networks (KANs) to transform latent features, circumventing the need for paired hazy-clean data. It introduces a Dual-GR-KAN Transformer within a ViT-inspired generator and employs adversarial training plus patch-wise contrastive learning to enforce semantic and structural fidelity. The approach achieves state-of-the-art performance on RESIDE-derived benchmarks with notably lower computational cost, and its variants offer favorable speed-accuracy trade-offs. The work demonstrates the potential of KANs for ill-posed image restoration tasks and provides code for reproducibility.

Abstract

This paper proposes an innovative framework for Unsupervised Image Dehazing via Kolmogorov-Arnold Transformation, termed UID-KAT. Image dehazing is recognized as a challenging and ill-posed vision task that requires complex transformations and interpretations in the feature space. Recent advancements have introduced Kolmogorov-Arnold Networks (KANs), inspired by the Kolmogorov-Arnold representation theorem, as promising alternatives to Multi-Layer Perceptrons (MLPs) since KANs can leverage their polynomial foundation to more efficiently approximate complex functions while requiring fewer layers than MLPs. Motivated by this potential, this paper explores the use of KANs combined with adversarial training and contrastive learning to model the intricate relationship between hazy and clear images. Adversarial training is employed due to its capacity in producing high-fidelity images, and contrastive learning promotes the model's emphasis on significant features while suppressing the influence of irrelevant information. The proposed UID-KAT framework is trained in an unsupervised setting to take advantage of the abundance of real-world data and address the challenge of preparing paired hazy/clean images. Experimental results show that UID-KAT achieves state-of-the-art dehazing performance across multiple datasets and scenarios, outperforming existing unpaired methods while reducing model complexity. The source code for this work is publicly available at https://github.com/tranleanh/uid-kat.

Paper Structure

This paper contains 16 sections, 17 equations, 8 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Single Image Dehazing
  4. Vision Transformers
  5. Contrastive Learning
  6. Kolmogorov-Arnold Networks (KANs)
  7. Methodology
  8. Overview
  9. Architectures
  10. Loss Functions
  11. Experiments
  12. Experimental Details
  13. Performance Evaluations
  14. Ablation Study
  15. Runtime
  16. ...and 1 more sections

Figures (8)

  • Figure 1: Trade-off between effectiveness (PSNR) and efficiency (MACs) on the SOTS-Outdoor and HSTS benchmarks. The proposed UID-KAT achieves competitive performance with low computational complexity.
  • Figure 2: The proposed UID-KAT framework.
  • Figure 3: Typical visual results on SOTS-Outdoor.
  • Figure 4: Typical visual results on HSTS.
  • Figure 5: Typical visual results on natural data.
  • ...and 3 more figures