SRGAN-CKAN: Super-Resolution with Kolmogorov-Arnold Networks
A new super-resolution framework, SRGAN-CKAN, integrates Convolutional Kolmogorov-Arnold Networks into an adversarial learning setting. It replaces linear local mappings with spline-based functional representations to enhance expressivity under minimal computational resources. The approach focuses on improving high-frequency texture reconstruction for large upscaling factors, addressing the ill-posed nature of single-image super-resolution.
Key facts
- SRGAN-CKAN is a hybrid super-resolution framework.
- It integrates Convolutional Kolmogorov-Arnold Networks (CKAN) into adversarial learning.
- The method reformulates convolution as a nonlinear patch-based transformation.
- It uses spline-based functional representations to replace linear local mappings.
- The goal is to reconstruct high-resolution images from low-resolution observations.
- The approach targets large upscaling factors where high-frequency details are degraded.
- The work is published on arXiv with ID 2605.01459.
- It contrasts with transformer and diffusion models that increase computational complexity.
Entities
Institutions
- arXiv