Latent Degradation Representation Constraint for Single Image Deraining
Yuhong He, Long Peng, Lu Wang, Jun Cheng
TL;DR
This work tackles single image deraining under diverse rain patterns by learning an explicit latent degradation representation. It introduces LDRCNet, featuring a Direction-Aware Encoder (DAEncoder) that uses deformable convolutions to extract direction-consistent degradation components $deg=ig\{deg_1,deg_2,deg_3\big\}$ from rainy input, and a constraint framework that supervises this learning by reconstructing the rainy image from a rain-free image and $deg$. An MSIBlock enables adaptive fusion of the learned degradation with decoder features in a U-Net deraining network, promoting robust removal of rain while preserving textures. The model is trained with a dual loss $\,\mathcal{L}_{\text{total}} = \lambda_1\mathcal{L}_D + \lambda_2\mathcal{L}_C$, enabling explicit degradation guidance and effective interaction between degradation and features. Extensive experiments on synthetic and real datasets show state-of-the-art performance, demonstrating strong generalization and practical impact for real-world deraining tasks.
Abstract
Since rain streaks show a variety of shapes and directions, learning the degradation representation is extremely challenging for single image deraining. Existing methods are mainly targeted at designing complicated modules to implicitly learn latent degradation representation from coupled rainy images. This way, it is hard to decouple the content-independent degradation representation due to the lack of explicit constraint, resulting in over- or under-enhancement problems. To tackle this issue, we propose a novel Latent Degradation Representation Constraint Network (LDRCNet) that consists of Direction-Aware Encoder (DAEncoder), UNet Deraining Network, and Multi-Scale Interaction Block (MSIBlock). Specifically, the DAEncoder is proposed to adaptively extract latent degradation representation by using the deformable convolutions to exploit the direction consistency of rain streaks. Next, a constraint loss is introduced to explicitly constraint the degradation representation learning during training. Last, we propose an MSIBlock to fuse with the learned degradation representation and decoder features of the deraining network for adaptive information interaction, which enables the deraining network to remove various complicated rainy patterns and reconstruct image details. Experimental results on synthetic and real datasets demonstrate that our method achieves new state-of-the-art performance.