Images captured in rainy days are often obscured by rain streaks which affect the accuracy of object detection, vehicle and pedestrian recognition. It is hard to restore the texture and color information of the de-rained image by some conventional rain removal algorithms. In order to address the problem, we propose a novel two stage conditional generative adversarial network (TS-CGAN), in which the generator network of the TS-CGAN contains two stage frameworks to better gradually remove rain streaks. In addition, compared with previous neural networks on the rain removal work, patch-GAN discriminator of the TS-CGAN can able to encourage generator to adversely produce satisfactory visual clean images that can solve the artifact problem. Extensive experiments on synthetic and real-world datasets demonstrate that the proposed method significantly outperforms recent state-of-the-art algorithms in terms of qualitative and quantitative measurement.

在雨天拍摄的图像通常会被雨条纹遮盖，这会影响物体检测，车辆和行人识别的准确性。通过一些常规的雨水去除算法很难恢复排水图像的纹理和颜色信息。为了解决该问题，我们提出了一种新颖的两阶段条件生成对抗网络（TS-CGAN），其中TS-CGAN的生成器网络包含两个阶段框架以更好地逐渐消除雨水条纹。此外，与以前的除雨工作神经网络相比，TS-CGAN的patch-GAN鉴别器可以鼓励发生器不利地产生令人满意的视觉清洁图像，从而解决伪影问题。