RASR: Retrieval-Augmented Super Resolution for Practical Reference-based Image Restoration

A new paradigm called Retrieval-Augmented Super Resolution (RASR) has been introduced to enhance reference-based image restoration by automatically retrieving high-resolution images from a database based on low-quality inputs. This approach aims to improve texture fidelity and visual realism without the need for manually curated target-reference image pairs, addressing a significant limitation of existing methods.

The development of RASR is significant as it enables scalable and flexible reference-based super resolution, making it practical for real-world applications such as enhancing mobile photos in specific environments like zoos or museums, where relevant reference data can be easily collected.