English

Generalized Real-World Super-Resolution through Adversarial Robustness

Image and Video Processing 2021-08-27 v1 Computer Vision and Pattern Recognition Machine Learning

Abstract

Real-world Super-Resolution (SR) has been traditionally tackled by first learning a specific degradation model that resembles the noise and corruption artifacts in low-resolution imagery. Thus, current methods lack generalization and lose their accuracy when tested on unseen types of corruption. In contrast to the traditional proposal, we present Robust Super-Resolution (RSR), a method that leverages the generalization capability of adversarial attacks to tackle real-world SR. Our novel framework poses a paradigm shift in the development of real-world SR methods. Instead of learning a dataset-specific degradation, we employ adversarial attacks to create difficult examples that target the model's weaknesses. Afterward, we use these adversarial examples during training to improve our model's capacity to process noisy inputs. We perform extensive experimentation on synthetic and real-world images and empirically demonstrate that our RSR method generalizes well across datasets without re-training for specific noise priors. By using a single robust model, we outperform state-of-the-art specialized methods on real-world benchmarks.

Keywords

Cite

@article{arxiv.2108.11505,
  title  = {Generalized Real-World Super-Resolution through Adversarial Robustness},
  author = {Angela Castillo and María Escobar and Juan C. Pérez and Andrés Romero and Radu Timofte and Luc Van Gool and Pablo Arbeláez},
  journal= {arXiv preprint arXiv:2108.11505},
  year   = {2021}
}

Comments

ICCV Workshops, 2021

R2 v1 2026-06-24T05:25:32.747Z