Organizations: School of Information Engineering, Guangdong University of Technology, Guangzhou, 510006, China · School of Computer Science and Engineering, Nanjing University of Science and Technology
Image deraining plays a pivotal role in low-level computer vision, serving as a prerequisite for robust outdoor surveillance and autonomous driving systems. While deep learning paradigms have achieved remarkable success in firmly aligned settings, they often suffer from severe performance degradation when generalized to unseen Out-of-Distribution (OOD) scenarios. This failure stems primarily from the significant domain discrepancy between synthetic training datasets and the complex physical dynamics of real-world rain. To address these challenges, this paper proposes a pioneering cross-scenario deraining adaptation framework. Diverging from conventional approaches, our method obviates the requirements for paired rainy observations in the target domain, leveraging exclusively rain-free background images. We design a Superpixel Generation (Sup-Gen) module to extract stable structural priors from the source domain using Simple Linear Iterative Clustering. Subsequently, a Resolution-adaptive Fusion strategy is introduced to align these source structures with target backgrounds through texture similarity, ensuring the synthesis of diverse and realistic pseudo-data. Finally, we implement a pseudo-label re-Synthesize mechanism that employs multi-stage noise generation to simulate realistic rain streaks. This framework functions as a versatile plug-and-play module capable of seamless integration into arbitrary deraining architectures. Extensive experiments on state-of-the-art models demonstrate that our approach yields remarkable PSNR gains of up to 32% to 59% in OOD domains while significantly accelerating training convergence.
Figures & tables
Fig. 1: The challenge of removing raindrop across various Out-of-distribution (OOD) scenarios. We first use the Rain200L dataset to train the NeRD-Rain, DRSformer, and FADformer methods respectively. In order to implement the evaluation on Out-of-distribution(OOD) scenarios, we use Rain200L as the source domain data, and Rain200H, DID, and DDN as the OOD domains for testing. Empirical results demonstrate that while the model achieves exemplary deraining performance on the source domain (Rain200L), the performance of state-of-the-art derain models deteriorates significantly when applied to unseen OOD domains. Thus, we call for a solid derain pipeline is required to better handle unknown scene conditions.
Fig. 2: An illustration of our proposed method: (A) Superpixel Generation: This module takes the source domain rain-free image Isource as input. Utilizing a superpixel segmentation algorithm, it parses the image into a collection of superpixel blocks rich in structural and textural information, denoted as Itargetsuperpixel . (B) Resolution-adaptive Fusion: This module performs local matching between the extracted superpixel blocks Itargetsuperpixel and the target domain’s pseudo rain-free image Itargetpsdo . It locates optimal integration regions Pp∗ based on semantic consistency and texture similarity(MSE). By employing a random mask matrix Mrandom for proportional extraction and applying an α -fusion strategy with a fusion coefficient of β , it generates an information-enhanced target image, Itargetfuse . (C) Pseudo-label Re-Synthesis: To synthesize high-fidelity pseudo rain streak layers, the module implements a three-stage degradation process comprising Salt-and-Pepper noise ( Xs ), Gaussian blurring ( Xg ), and motion blurring ( Xm ). This streak layer is superimposed onto the luminance channel of Itargetfuse using α fusion to create the corresponding rainy image, Iinputfuse . With the synergistic operation of these three modules, a final set of pseudo-paired samples (Iinputfuse,Itargetfuse) is obtained, significantly enhancing the model’s generalization toward out-of-distribution (OOD) domains.
Fig. 3: The framework of Superpixel Generation, the process initiates by incorporating spatial positional information (x,y) into the image data (l,a,b) . Subsequently, cluster centroids (li,ai,bi,xi,yi) are initialized based on predefined parameters.
Fig. 4: The framework of Resolution-Adaptive Fusion. We propose a novel fusion paradigm driven by bidirectional region matching and adaptive fusion mechanisms. This approach simultaneously incorporates structural priors from the source domain while preserving the intrinsic background distribution of the target domain, thereby facilitating subsequent domain adaptation tasks.
Fig. 5: Pseudo-label Re-Synthesis introduces a pseudo-rain streak generation methodology grounded in multi-stage noise synthesis. By sequentially executing noise initialization, morphological transformation, and luminance channel fusion within the YUV color space, the proposed method achieves the realistic simulation of rain effects on rain-free images within the target domain.
Fig. 6: Derained results on the Rain200L dataset. Compared with the derained results without our method, our method recovers a high-quality image with clearer details. Zooming in the figures offers a better view at the deraining capability.
Baseline
Testing Set: Rain200L (OOD)
PSNR ↑
SSIM ↑
LPIPS ↓
FSIM ↑
NIQE ↓
PI ↓
NeRD-Rain [ 19 ] w/o ours
25.44
0.8372
0.2483
0.8872
4.1645
2.7664
NeRD-Rain [ 19 ] w/ ours
33.67 (32% ↑ )
0.9495 (13% ↑ )
0.1000 (60% ↑ )
0.9559 (8% ↑ )
3.2822 (21% ↑ )
2.3327 (16% ↑ )
DRSformer [ 20 ] w/o ours
28.35
0.8787
0.1969
0.9068
3.6658
2.5371
DRSformer [ 20 ] w/ ours
32.68 (15% ↑ )
0.9478 (7% ↑ )
0.0849 (57% ↑ )
0.9547 (5% ↑ )
3.2469 (11% ↑ )
2.3505 (11% ↑ )
FADformer [ 21 ] w/o ours
27.18
0.8535
0.2375
0.8915
4.1380
2.7486
TABLE I: Quantitative Comparison on OOD Dataset.
Baseline
Variant
Best Ep.
PSNR ↑
SSIM ↑
NeRD-Rain
random
84
27.70
0.8779
Ours
15 ( 5.6 × )
33.67
0.9495
DRSformer
random
26
24.52
0.8515
Ours
6 ( 4.3 × )
32.68
0.9478
FADformer
random
150
28.70
0.8894
Ours
101 ( 1.5 × )
33.85
0.9550
TABLE II: Comparison of Training Efficiency and Performance
Fig. 7: The ablation study of our algorithm on different baseline methods. Our algorithm has greatly enhanced the convergence speed and generalization capability on cross-scenario environments.
α
0.8
0.6
0.4
0.2
PSNR ↑
32.84
32.83
32.59
33.08
SSIM ↑
0.9427
0.9420
0.9412
0.9460
TABLE III: Ablation Study on Resolution-adaptive Fusion Scale ( α )
Unsupervised deraining has attracted attention for its ability to learn the real-world distribution of rain without paired supervision. However, the lack of strong constraints makes it difficult for the network to converge, especially with the complex diversity of rain degradation. A key motivation is that high-quality deraining results occasionally emerge during training, which can be leveraged to guide the optimization process. To overcome these challenges, we introduce RGSUD (Reward-Guided Self-Reinforcement Unsupervised Image Deraining), comprising two key stages: reward recycling and self-reinforcement (SR) training. For the former stage, we propose an Image Quality Assessment (IQA)-based dynamic reward recycling mechanism that selects optimal derained outputs during training and continuously collects high-quality deraining images. In latter stage, we incorporate these rewards into the model's optimization process, constraining the optimization space and improving alignment between derained outputs and clean images. By leveraging IQA-based self-reinforced loss and dynamically updated rewards, we enhance the quality of synthesized pseudo-paired data and stabilize the optimization. Extensive experiments demonstrate that our method achieves SOTA performance across multiple datasets, including paired synthetic, paired real, and unpaired real images, outperforming existing unsupervised deraining approaches in both subjective and objective IQA metrics. Additionally, we show that the self-reinforcement strategy is adaptable to other unsupervised deraining methods and our deraining framework demonstrates strong generalization across existing supervised deraining networks.
Yinghao Chen, Yeying Jin, Xiang Chen +3
1National University of Defense Technology · National University of Singapore · 3Nanjing University of Science and Technology +2
Despite significant progress has been made in image deraining, we note that most existing methods are often developed for only specific types of rain degradation and fail to generalize across diverse real-world rainy scenes. How to effectively model different rain degradations within a universal framework is important for real-world image deraining. In this paper, we propose UniRain, an effective unified image deraining framework capable of restoring images degraded by rain streak and raindrop under both daytime and nighttime conditions. To better enhance unified model generalization, we construct an intelligent retrieval augmented generation (RAG)-based dataset distillation pipeline that selects high-quality training samples from all public deraining datasets for better mixed training. Furthermore, we incorporate a simple yet effective multi-objective reweighted optimization strategy into the asymmetric mixture-of-experts (MoE) architecture to facilitate consistent performance and improve robustness across diverse scenes. Extensive experiments show that our framework performs favorably against the state-of-the-art models on our proposed benchmarks and multiple public datasets.
Qianfeng Yang, Qiyuan Guan, Xiang Chen +3
Dalian Polytechnic University · Nanjing University of Science and Technology
While deep learning has advanced single-image deraining, existing models suffer from a fundamental limitation: they employ a static inference paradigm that fails to adapt to the complex, coupled degradations (e.g., noise artifacts, blur, and color deviation) of real-world rain. Consequently, restored images often exhibit residual artifacts and inconsistent perceptual quality. In this work, we present Derain-Agent, a plug-and-play refinement framework that transitions deraining from static processing to dynamic, agent-based restoration. Derain-Agent equips a base deraining model with two core capabilities: 1) a Planning Network that intelligently schedules an optimal sequence of restoration tools for each instance, and 2) a Strength Modulation mechanism that applies these tools with spatially adaptive intensity. This design enables precise, region-specific correction of residual errors without the prohibitive cost of iterative search. Our method demonstrates strong generalization, consistently boosting the performance of state-of-the-art deraining models on both synthetic and real-world benchmarks.