AI-GAN: Asynchronous interactive generative adversarial network for single image rain removal

作者：

Highlights：

• We propose to solve the image deraining problem from the perspective of feature-wise disentanglement.

• An end-to-end learning deraining model Asynchronous Interactive Generative Adversarial Network (AI-GAN) is proposed.

• AI-GAN separates the background and rain images based on a two-branch architecture and achieves complementary optimization.

• AI-GAN outperforms state-of-the-art deraining methods and benefits a large range of multimedia applications.

摘要

•We propose to solve the image deraining problem from the perspective of feature-wise disentanglement.•An end-to-end learning deraining model Asynchronous Interactive Generative Adversarial Network (AI-GAN) is proposed.•AI-GAN separates the background and rain images based on a two-branch architecture and achieves complementary optimization.•AI-GAN outperforms state-of-the-art deraining methods and benefits a large range of multimedia applications.

论文关键词：Feature-wise disentanglement,Asynchronous and interactive,Single image deraining,Complementary adversarial training

论文评审过程：Received 27 January 2019, Revised 26 July 2019, Accepted 27 November 2019, Available online 6 December 2019, Version of Record 13 December 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.107143