RASR: Retrieval-Augmented Super Resolution for Practical Reference-based Image Restoration
Researchers have introduced Retrieval-Augmented Super Resolution (RASR), a novel approach to image restoration that addresses the limitations of existing reference-based methods. Unlike previous techniques requiring manually paired images, RASR automatically retrieves relevant high-resolution reference images from a database, making it more practical for real-world applications like enhancing mobile photos. The team also developed RASRNet, a baseline model that combines a semantic retriever with a diffusion-based generator, and created RASR-Flickr30, the first benchmark dataset for this task. AI
IMPACT This research could lead to more practical and effective image enhancement tools for consumer devices.