DDTNet: Degradation Disentanglement and Transfer Network for Test-Time All-in-One De-weathering Adaptation
Researchers have developed the Degradation Disentanglement and Transfer Network (DDTNet), a novel approach for improving all-in-one adverse weather image restoration models. DDTNet focuses on disentangling degradation patterns from images and transferring them to clean images, thereby creating domain-adaptive training data. This method aims to overcome the performance compromises often seen in single models designed to handle multiple weather conditions like rain, haze, and snow, especially when there's a domain gap between training and testing data. The core of DDTNet, the Degradation Disentanglement Module (DDM) with Degradation Coupled Attention (DCA), effectively captures weather-specific features for improved adaptability. AI
IMPACT Enhances adaptability of image restoration models across diverse weather conditions and domains.