Single image deraining is a highly ill-posed problem. Existing deep neural network-based algorithms usually use larger deep models to solve this problem, which is less effective and efficient. In this paper, we propose a deep neural network based on feature pyramid to solve image deraining. Our algorithm is motivated that the features at different pyramid levels share similar structures. Based on this property, we develop an effective deep neural network, where the deep models at different feature pyramid levels share the same weight parameters. In addition, we further develop a multi-stream dilation convolution to deal with complex rainy streaks. To preserve the image detail, we develop dense connections that can maintain important features from different levels. Our algorithm is trained in an end-to-end manner. Quantitative and qualitative experimental results demonstrate that the proposed method performs favorably against state-of-the-art deraining methods in terms of accuracy as well as model sizes. The source code and dataset will be available at https://supercong94.wixsite.com/supercong94.

中文翻译：

---

通过深度共享金字塔网络去除单个图像

单幅图像去雨是一个高度不适定的问题。现有的基于深度神经网络的算法通常使用更大的深度模型来解决这个问题，效率较低。在本文中，我们提出了一种基于特征金字塔的深度神经网络来解决图像去雨问题。我们的算法的动机是不同金字塔级别的特征共享相似的结构。基于这个特性，我们开发了一个有效的深度神经网络，其中不同特征金字塔级别的深度模型共享相同的权重参数。此外，我们进一步开发了一个多流扩张卷积来处理复杂的雨条纹。为了保留图像细节，我们开发了可以从不同层次维护重要特征的密集连接。我们的算法以端到端的方式进行训练。定量和定性实验结果表明，所提出的方法在准确性和模型大小方面均优于最先进的去雨方法。源代码和数据集将在 https://supercong94.wixsite.com/supercong94 上提供。