Original title: Efficient All-in-One Weather Restoration using Spectral Harmonization
Paula Garrido-Mellado, Daniel Feijoo, Yuning Cui, Alvaro Garcia +1 · September 2026
Bad weather like rain, haze, and snow ruins image quality, making pictures hard for people and physical AI to use. Current fixing methods cost too much computer power. They struggle to handle different kinds of weather damage and cannot easily process high-resolution images. To solve this, researchers built an all-in-one method called Frequency ReconstructionRebuilding an image by separating it into high-frequency (sharp details) and low-frequency (smooth areas) components. via Spectral HarmonizationAdjusting different frequency parts of an image so they work together smoothly during restoration., or FReSH-IR. It works by splitting image features into high-frequency parts, which show sharp edges, and low-frequency parts, which show smooth colors. It does this across different image scales inside an encoder-decoder network, combining frequency and spatial processing with special Fourier-based connections. This setup keeps fine spatial details while capturing needed frequency information. It matches the restoration quality of transformer-based models while using 80% fewer parameters and operationsThe numbers that define how large a model is and how much computing it requires to run..
You're packing for a vacation where the forecast calls for sun, rain, and snow all in one day. Your suitcase is tiny, so you can't bring separate outfits for each weather. In this paper, the image system faces similar constraints—it must clean up photos from different weather types without having a lot of computing resources to spare. It does this by splitting the image into different frequency layers and processing them efficiently, much like choosing clothes that adapt to various conditions.
When rain or snow blurs a camera feed, this method can clean it up without needing a powerful computer. This makes reliable vision possible in phones or drones where space and battery are tight.
The abstract doesn't specify limitations, but it focuses on efficiency for constrained-resource systems and may not address performance in all weather conditions or on extremely high-resolution images without validation.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Efficient All-in-One Weather Restoration using Spectral Harmonization — Paula Garrido-Mellado, Daniel Feijoo, Yuning Cui et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.