Abstract
Visible and Infrared image fusion(VIF) methods in different scenarios have improved significantly over the recent years. Yet, the existing methods rarely take severe image degradation caused by weather conditions into account, and a few of those that do, cannot cope with diverse types and severity of degradation. The progress in the development of robust all-weather VIF algorithms has been hampered by the scarcity of paired degraded image training data. In this paper, we report a new Visible-infrared weather degradation dataset, labeled with degradation types and severity, which has been compiled to facilitate the training of a VIF model that is robust to images degraded by adverse weather conditions. We propose a framework that incorporates a refined CLIP model to guide the VIF model capable of automatically detecting and grading image degradation to control the fusion process. The image restoration guidance is prompt-based. This approach eliminates the need for manual intervention during the inference phase. This innovative solution and the dataset advance the VIF research towards robust open world fusion. The results of extensive experiments demonstrate the efficacy and superiority of our proposed method. Our code and dataset will be released soon.