Attention-based Adaptive Selection of Operations for Image Restoration in the Presence of Unknown Combined Distortions

This repository contains the code for the following paper:

Masanori Suganuma, Xing Liu, Takayuki Okatani, "Attention-based Adaptive Selection of Operations for Image Restoration in the Presence of Unknown Combined Distortions," CVPR, 2019. [arXiv]

If you find this work useful in your research, please cite:

@inproceedings{suganumaCVPR2019,
    Author = {M. Suganuma and X. Liu and T. Okatani},
    Title = {Attention-based Adaptive Selection of Operations for Image Restoration in the Presence of Unknown Combined Distortions},
    Booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    Year = {2019}
}

Sample results on image restoration:

example

Sample results on object detection:

example

Requirement

Usage

Train a model on the dataset proposed by RL-Restore

python main.py -m mix -g 1

When you use the multiple GPUs, please specify the number of gpus by -g option (default:1)

Dataset

The dataset used in RL-Restore is available here. To generate the training dataset, please run data/train/generate_train.m in the repository and put the generated file (train.h5) to dataset/train/ in your computer.

Train a model on your own dataset

python main.py -m yourdata -g 1

Test

Put the trained model (XXXX.pth) to Trained_model/, and run the following code:

python test.py -m mix -g 1