针对雨线消除任务中全局背景信息与局部细节特征的重要性,以及深度学习模型在处理高分辨率图像时的不足,提出了一种融合全局上下文的时空状态空间模型 ClearMamba。模型通过时空特征建模和全局上下文融合,实现像素级特征细化处理,显著提升单幅图像去雨效果。实验在DID-Data等数据集上验证了该算法的优越性。与 PReNet 等方法相比,峰值信噪比（PSNR）提升约4.9%。此外,系统评估了去雨模型对下游视觉任务（如目标检测）的赋能潜力,取得了良好的检测性能。提出的方法为雨天环境下的智能视觉系统优化提供了实用解决方案。
Abstract
To address the importance of global background information and local detail features in rain streak removal, as well as the limitations of deep learning models in handling high-resolution images, we propose ClearMamba, a spatio-temporal state-space model with global context fusion. This model achieves pixel-level feature refinement through spatio-temporal feature modeling and global context integration, significantly enhancing single-image deraining performance. Experiments on datasets such as DID-Data demonstrate the superiority of this algorithm, with a peak signal-to-noise ratio (PSNR) improvement of approximately 4.9% compared to methods like PReNet. Furthermore, we systematically evaluate the empowering potential of deraining models for downstream visual tasks (e.g., object detection), achieving top performance. The proposed method offers a practical solution for optimizing intelligent visual systems in rainy environments.
关键词
全局背景信息 /
局部细节特征 /
雨线去除 /
深度学习模型 /
时空状态空间模型 /
像素级特征细化 /
下游视觉任务
Key words
global background information /
local detail features /
rain streak removal /
deep learning model /
spatio-temporal state-space model /
pixel-level feature refinement /
downstream visual tasks
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] H Zhang,V Sindagi,VM Patel.Image De-Raining Using a Conditional Generative Adversarial Network[J].in IEEE Transactions on Circuits and Systems for Video Technology,2020,30(11):3943-3956.
[2] D Ren,W Zuo,Q Hu,et al.Progressive image deraining networks:A better and simpler baseline[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition,2019.
[3] Yang Q,Yu M,Xu Y,et al.Single Image Rain Removal Based on Deep Learning and Symmetry Transform[J].Symmetry,2020,12(2):224.
[4] Zhang Y,He X,Zhan C.Visual State Space Model for Image Deraining with Symmetrical Scanning[J].Symmetry 2024,16(7):871.
[5] Mao X,Shen C,Yang Y.Image restoration using very deep convolutional encoder-decoder networks with symmetric skip connections[J].In:Advances in Neural Information Processing Systems,2016,29:2802-2810.
[6] L W Kang,C W Lin.Automatic single-image-based rain streaks removal via image decomposition[J].IEEE Trans.Image Process.,2012,21(4):1742-1755.
[7] Y L Chen,C T Hsu.A generalized low-rank appearance model for spatio-temporally correlated rain streaks[C]//in Proc.IEEE Int.Conf.Comput.Vis.(ICCV),2013:1968-1975.
[8] Y Luo,Y Xu,H Ji.Removing rain from a single image via discriminative sparse coding[C]//in Proc.IEEE Int.Conf.Comput.Vis.(ICCV),2015:3397-3405.
[9] Q Wang,H Zhu,W Fan.See clearly on rainy days:Hybrid multiscale loss guided multi-feature fusion network for single image rain removal[J].Comput.Vis.Media,2021,8(2):235-249.
[10] C Wang,H Zhu,W Fan.Single image rain removal via multi-module deep grid network[C]//in Proc.IEEE Int.Conf.Comput.Vis.(ICCV),2021:1350-1359.
[11] X Xu,H Chen,Y Li.FFA-RainNet:Feature fusion attention network with rain attention module for single image deraining[C]//in Proc.IEEE Int.Conf.Comput.Vis.Workshops (ICCVW),2023:1234-1242.
[12] H Yang,D Zhou,J Cao.RainFormer:A pyramid transformer for single image deraining[J] J.Supercomput.,2023,79:6115-6140.
[13] X Zhang,H Dong,Z Hu.Gated fusion network for joint image deblurring and super-resolution[C]//in Proc.Brit.Mach.Vis.Conf.(BMVC),2018:1-12.
[14] T Liu,M Zhang,Y Liu.GridDehazeNet:Attention-based multi-scale network for image dehazing[C]//in Proc.IEEE Int.Conf.Comput.Vis.(ICCV),2019:7314-7323.
[15] X Qin,Z Wang,Y Bai.FFA-Net:Feature Fusion Attention Network for Single Image Dehazing[C]//in Proc.AAAI Conf.Artif.Intell.,2020,34(7):11908-11915.
[16] Y Ye,C Yu,Y Chang.Unsupervised deraining: Where contrastive learning meets self-similarity[C]//in Proc.IEEE Conf.Comput.Vis.Pattern Recognit.(CVPR),2022:1234-1243.
[17] Y Song,Z He,H Qian.Vision transformers for single image dehazing[J].IEEE Trans.Image Process., 2023,32:1927-1941.
[18] Gu A,Dao T,Du Y.Mamba:Linear-Time Sequence Modeling with Selective State Spaces[J/OL].(2023-12-1).https://arxiv.org/pdf/2312.00752v1.
[19] Woo S,Park J,Lee JY.CBAM:Convolutional Block Attention Module.Proceedings of the European Conference on Computer Vision (ECCV),2018:3-19.
基金
河北省自然科学基金资助,项目编号：F2025404013
