Search papers, labs, and topics across Lattice.
This paper introduces DDTNet, a novel framework for all-in-one adverse weather image restoration that addresses the limitations of existing models by focusing on degradation transfer rather than direct content recovery. By employing a Degradation Disentanglement Module (DDM) with Degradation Coupled Attention (DCA), DDTNet effectively disentangles degradation patterns from target-domain images and transfers them to clean images from the source domain, creating domain-adaptive training pairs. Experimental results show that DDTNet significantly enhances the performance of restoration models across various weather conditions, outperforming existing methods in real-world scenarios.
DDTNet reveals that focusing on degradation transfer can dramatically improve image restoration performance across diverse weather conditions.
All-in-one adverse weather image restoration aims to remove multiple degradations, such as rain, haze, and snow, using a single unified model. Despite their broad applicability, existing methods typically compromise performance, delivering balanced but suboptimal results for individual degradation types. This issue becomes more pronounced when a domain gap exists between training and testing data. Motivated by the observation that modeling degradation patterns is more feasible than recovering clean content, we propose the Degradation Disentanglement and Transfer Network (DDTNet), which focuses specifically on degradation transfer. By disentangling degradation patterns from target-domain degraded images and transferring them to source domain clean images, DDTNet generates domain-adaptive paired training data. These pairs are then used to fine-tune restoration models, significantly enhancing their adaptability across diverse weather conditions and domains. The core of DDTNet is the Degradation Disentanglement Module (DDM), which comprises Degradation Coupled Attention (DCA) to capture both general and weather-specific features, thereby enabling effective disentanglement and transfer of degradation patterns. Experimental results demonstrate that DDTNet significantly and consistently improves existing all-in-one models across real-world deraining, desnowing, and dehazing datasets.
