Crafting Training Degradation Distribution for the Accuracy-Generalization Trade-off in Real-World Super-Resolution

TL;DR

This paper presents a method to optimize training degradation distributions for real-world super-resolution, balancing generalization and accuracy by using a small reference set and a novel distribution crafting approach.

Contribution

It introduces a new technique to craft training degradation distributions based on binned representations and Fréchet distance, improving test performance while maintaining generalization.

Findings

Significant improvement in test image performance.

Preservation of generalization capabilities.

Effective trade-off between accuracy and generalization.

Abstract

Super-resolution (SR) techniques designed for real-world applications commonly encounter two primary challenges: generalization performance and restoration accuracy. We demonstrate that when methods are trained using complex, large-range degradations to enhance generalization, a decline in accuracy is inevitable. However, since the degradation in a certain real-world applications typically exhibits a limited variation range, it becomes feasible to strike a trade-off between generalization performance and testing accuracy within this scope. In this work, we introduce a novel approach to craft training degradation distributions using a small set of reference images. Our strategy is founded upon the binned representation of the degradation space and the Fr\'echet distance between degradation distributions. Our results indicate that the proposed technique significantly improves the performance of test images while preserving generalization capabilities in real-world applications.