【论文阅读】Single Image Deraining：A Comprehensive Benchmark Analysis

CVPR2019论文。论文没有细看，重点在于关注去雨模型中常用的三种建模方式。

• rain steak

  $$\mathbf{R}_{s}=\mathbf{B}+\mathbf{S}$$

• rain drop

  $$\mathbf{R}_{d}=(1-\mathbf{M}) \odot \mathbf{B}+\mathbf{D}$$

M is a binary mask and ⊙ means element-wise multiplication. In the mask, a pixel x is part of a raindrop region if
M(x) = 1, and otherwise belongs to the background.

• rain and mist $$\mathbf{R}_{m}=\mathbf{B} \odot t+A(1-t)+\mathbf{S}$$

where S is the rain streak component; t and A are the
transmission map and atmospheric light that determines the
fog/mist component, respectively.

三种不同的建模方式决定了去雨网络不同。未来我应该更关注rain drop和mist混合的情况。