WTNet: A Weather Transfer Network for Domain-Adaptive All-In-One Adverse Weather Image Restoration


Si-Yu Huang (National Yang Ming Chiao Tung University), Fu-Jen Tsai (National Tsing Hua University), Chia-Wen Lin (National Tsing Hua University), Yen-Yu Lin (National Yang Ming Chiao Tung University)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)

Abstract

All-in-one adverse weather image restoration has attracted increasing attention due to its ability to recover high-quality images in a unified model. However, existing methods often suffer from significant performance drops due to the domain gap between training and testing images. Moreover, they typically yield only balanced, rather than optimal, performance across different weather conditions, especially when compared to models trained in a weather-specific manner. To solve this problem, we propose a novel Weather Transfer Network (WTNet), which fine-tunes all-in-one models to enhance their performance during testing. As paired degraded-clean images are unavailable during testing, WTNet transfers degraded patterns from unseen target domains to the source-domain clean images, thereby constructing domain-adaptive fine-tuning sets for effective domain adaptation. Additionally, by leveraging these fine-tuning sets, all-in-one models can be dynamically adapted to weather-specific or mixed weather models based on the degradation patterns observed during testing. Experimental results demonstrate that WTNet can significantly enhance state-of-the-art all-in-one models on benchmark real-world image deraining, desnowing, and dehazing datasets. The source code is available at https://github.com/stellaahuang/WTNet.

Citation

@inproceedings{Huang_2025_BMVC,
author    = {Si-Yu Huang and Fu-Jen Tsai and Chia-Wen Lin and Yen-Yu Lin},
title     = {WTNet: A Weather Transfer Network for Domain-Adaptive All-In-One Adverse Weather Image Restoration},
booktitle = {36th British Machine Vision Conference 2025, {BMVC} 2025, Sheffield, UK, November 24-27, 2025},
publisher = {BMVA},
year      = {2025},
url       = {https://bmva-archive.org.uk/bmvc/2025/assets/papers/Paper_150/paper.pdf}
}


Copyright © 2025 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection