Abstract:Removing adverse weather conditions like rain, fog, and snow from images is a challenging problem. Although the current recovery algorithms targeting a specific condition have made impressive progress, it is not flexible enough to deal with various degradation types. We propose an efficient and compact image restoration network named DAN-Net (Degradation-Adaptive Neural Network) to address this problem, which consists of multiple compact expert networks with one adaptive gated neural. A single expert network efficiently addresses specific degradation in nasty winter scenes relying on the compact architecture and three novel components. Based on the Mixture of Experts strategy, DAN-Net captures degradation information from each input image to adaptively modulate the outputs of task-specific expert networks to remove various adverse winter weather conditions. Specifically, it adopts a lightweight Adaptive Gated Neural Network to estimate gated attention maps of the input image, while different task-specific experts with the same topology are jointly dispatched to process the degraded image. Such novel image restoration pipeline handles different types of severe weather scenes effectively and efficiently. It also enjoys the benefit of coordinate boosting in which the whole network outperforms each expert trained without coordination. Extensive experiments demonstrate that the presented manner outperforms the state-of-the-art single-task methods on image quality and has better inference efficiency. Furthermore, we have collected the first real-world winter scenes dataset to evaluate winter image restoration methods, which contains various hazy and snowy images snapped in winter. Both the dataset and source code will be publicly available.

Image Desnowing via Deep Invertible Separation

Automatical Enhancement and Denoising of Extremely Low-light Images.

Spatial-Adaptive Network for Single Image Denoising

Context-aware Coarse-to-fine Network for Single Image Desnowing

Star-Net: Improving Single Image Desnowing Model With More Efficient Connection and Diverse Feature Interaction

Towards Real-time High-Definition Image Snow Removal: Efficient Pyramid Network with Asymmetrical Encoder-decoder Architecture

DesnowNet: Context-Aware Deep Network for Snow Removal

Lightweight image de-snowing: A better trade-off between network capacity and performance

Interactive Separation Network for Image Inpainting

Depth prior-based stable tensor decomposition for video snow removal

Self-Supervised Learning Hyperspectral Image Denoiser with Separated Spectral-Spatial Feature Extraction.

Deep Dense Multi-scale Network for Snow Removal Using Semantic and Geometric Priors

Semi-Supervised Video Desnowing Network via Temporal Decoupling Experts and Distribution-Driven Contrastive Regularization

Towards Efficient Single Image Dehazing and Desnowing

Snow Mask Guided Adaptive Residual Network for image snow removal

SnowFormer: Context Interaction Transformer with Scale-awareness for Single Image Desnowing

Video Desnowing and Deraining via Saliency and Dual Adaptive Spatiotemporal Filtering

A new Image Decomposition Approach using Pixel-wise Analysis Sparsity Model

SGNet: Efficient Snow Removal Deep Network with a Global Windowing Transformer

Distinguishing Cloud and Snow in Satellite Images Via Deep Convolutional Network

Extraction of snow cover from high-resolution remote sensing imagery using deep learning on a small dataset