Single Image Deraining Using a Recurrent Dual-Attention-Residual Ensemble Network

Abstract

Single image deraining is an ill-posed inverse problem due to the presence of non-uniform rain shapes, directions, and densities in images. In this paper, we propose a Single Image Deraining Using a Recurrent Dual-Attention-Residual Ensemble Network(RDARNet). Extensive experiments demonstrate that the effect of removing rain and restoring texture details is greatly improved.

Dataset

Synthetic Datasets

Datasets	train	test
Rain100L	200	100
Rain100H	1800	100
Rain800	700	100
Rain12		12

Pre-trained Model

We note that these models is trained on NVIDIA GeForce RTX2080Ti:

Datasets	Pre-trained model
Rain100H	Rain100H model

Requirements

• python 3.6.8
• opencv 4.1.2
• pyotrch 1.0.0

Usages

• Clone this repo

   $ git clone https://github.com/rainbowH/RDARENet
   $ cd RDARENet

• Test

   $ python RDARENet_test.py