FREDSR: Fourier Residual Efficient Diffusive GAN for Single Image Super Resolution

TL;DR

FREDSR is a novel GAN variant that leverages Fourier transforms and residual diffusion techniques to achieve high-performance single image super resolution with extreme parameter efficiency, optimized for specific datasets.

Contribution

Introduces FREDSR, a GAN model combining Fourier transforms and residual diffusion for efficient super resolution with fewer parameters and dataset-specific optimization.

Findings

Outperforms existing models on UHDSR4K dataset

Achieves high-quality 3x super resolution from low-resolution images

Uses only 37,000 parameters for real-time upscaling

Abstract

FREDSR is a GAN variant that aims to outperform traditional GAN models in specific tasks such as Single Image Super Resolution with extreme parameter efficiency at the cost of per-dataset generalizeability. FREDSR integrates fast Fourier transformation, residual prediction, diffusive discriminators, etc to achieve strong performance in comparisons to other models on the UHDSR4K dataset for Single Image 3x Super Resolution from 360p and 720p with only 37000 parameters. The model follows the characteristics of the given dataset, resulting in lower generalizeability but higher performance on tasks such as real time up-scaling.