Abstract:Existing unpaired image deraining approaches face challenges in accurately capture the distinguishing characteristics between the rainy and clean domains, resulting in residual degradation and color distortion within the reconstructed images. To this end, we propose an energy-informed diffusion model for unpaired photo-realistic image deraining (UPID-EDM). Initially, we delve into the intricate visual-language priors embedded within the contrastive language-image pre-training model (CLIP), and demonstrate that the CLIP priors aid in the discrimination of rainy and clean images. Furthermore, we introduce a dual-consistent energy function (DEF) that retains the rain-irrelevant characteristics while eliminating the rain-relevant features. This energy function is trained by the non-corresponding rainy and clean images. In addition, we employ the rain-relevance discarding energy function (RDEF) and the rain-irrelevance preserving energy function (RPEF) to direct the reverse sampling procedure of a pre-trained diffusion model, effectively removing the rain streaks while preserving the image contents. Extensive experiments demonstrate that our energy-informed model surpasses the existing unpaired learning approaches in terms of both supervised and no-reference metrics.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve several key challenges in **unpaired image deraining**. Specifically, existing methods have difficulties in accurately capturing the distinctive features between rainy - day and sunny - day images, resulting in problems such as residual degradation and color distortion in the reconstructed images. Therefore, this paper proposes a new energy - informed diffusion model (EDM) for the unpaired photo - realistic image deraining task (UPID - EDM). #### Main problems include: 1. **Inaccurate feature capture**: Existing unpaired deraining methods are difficult to accurately capture the distinctive features between rainy - day and sunny - day images, leading to a decline in the quality of reconstructed images. 2. **Residual degradation and color distortion**: Due to the lack of clear constraints, existing methods are prone to produce residual degradation and color distortion when processing real - world images. 3. **Data scarcity**: Precisely labeled deraining data is very scarce, which limits the application of supervised learning methods. To solve these problems, this paper proposes the following innovations: - **Introducing the Dual - consistent Energy Function (DEF)**: DEF is trained with non - corresponding rainy - day and sunny - day images and can retain image content while removing rain streaks. - **Utilizing the Contrastive Language - Image Pretraining model (CLIP)**: Extract visual - language prior knowledge through the CLIP model to help distinguish between rainy - day and sunny - day images. - **Learnable Domain - representation Prompts (LDP)**: LDP classifies rainy - day and sunny - day images through the binary cross - entropy loss function to ensure that the generated images conform to the sunny - day domain. - **Decomposing the energy function**: Decompose the energy function into two parts, which are respectively responsible for discarding rain - related features and retaining rain - unrelated features, thereby improving the quality and naturalness of the reconstructed images. Through these methods, the UPID - EDM model proposed in this paper shows superior deraining effects on multiple public datasets, especially achieving the best performance on both supervised and no - reference evaluation metrics. ### Summary The main goal of this paper is to solve the problems of feature capture, residual degradation, and color distortion in unpaired image deraining by introducing an energy - informed diffusion model, thereby achieving high - quality photo - realistic image deraining.

Unpaired Photo-realistic Image Deraining with Energy-informed Diffusion Model

Rethinking Real-world Image Deraining via An Unpaired Degradation-Conditioned Diffusion Model

Uncertainty-Aware Cascaded Dilation Filtering for High-Efficiency Deraining

Unpaired Deep Image Deraining Using Dual Contrastive Learning

DRD-Net: Detail-recovery Image Deraining via Context Aggregation Networks

Detail-recovery Image Deraining via Dual Sample-augmented Contrastive Learning

Enhanced Attentive Generative Adversarial Network for Single-Image Deraining.

EfficientDeRain+: Learning Uncertainty-Aware Filtering via RainMix Augmentation for High-Efficiency Deraining

Importing Diffusion and Re-Designed Backward Process for Image De-Raining

Single Image Deraining: From Model-Based to Data-Driven and Beyond

Physical Model Guided Deep Image Deraining

Universal Image Restoration with Text Prompt Diffusion

Real‐World Image Deraining Using Model‐Free Unsupervised Learning

Context-detail-aware United Network for Single Image Deraining

Image deraining via multi-level decomposition and empirical wavelet transform

Local and global knowledge distillation with direction-enhanced contrastive learning for single-image deraining

Latent Degradation Representation Constraint for Single Image Deraining

An Efficient Single Image De-Raining Model With Decoupled Deep Networks

Rainmer: Learning Multi-view Representations for Comprehensive Image Deraining and Beyond

Not Just Streaks: Towards Ground Truth for Single Image Deraining

Deep Single Image Deraining Via Estimating Transmission and Atmospheric Light in rainy Scenes