Home

Awesome

Encoder-Decoder Residual Network (IVIP-Lab)

This repository is our solution for NTIRE2019 Real Super-Resolution Challenge.

We, team IVIP-LAB, won the 9th PSNR and Top5 SSIM in the final phase of NTIRE2019 Real Super-resolution challenge.
Our paper will be published in CVPR 2019 Workshop. </i> [pdf] [poster]

Table of contents

<a id="network"></a>

Network Architecture

Encoder-Decoder Residual Network (EDRN)
Overview of EDRN

<a id="result"></a>

NTIRE2019 Real Super-resolution Challenge Results

Quantitative Results

MethodPSNR (dB)SSIMRuntime (s)
Baseline26.890.78--
EDRN (ours)28.790.8447.08

Qualitative Results

Qualitative_results

Classic single image super-resolution benchmark results

SISR_results

About Our Source Code & Trained Model

Dependencies

Our code is tested on Ubuntu 16.04 environment with an NVIDIA GTX 1080Ti GPU.

Dataset

Please download the dataset from here, and then put the downloaded dataset into ntire file.

Training

$ python main.py --save EDRN --save_results --save_models --model edrn --patch_size 128 --lr_decay 50 --n_GPUs 1 --chop

Test

Quick start (Demo) to reproduce our results. Please download our pretrained model from here.

$ python main.py --save EDRN --save_results --model edrn --n_GPUs 1 --chop --pre_train ../experiment/model/EDRN.pt --self_ensemble --test_only

Citation

If you find this work useful in your reseach, please cite our paper.

@InProceedings{Cheng_2019_CVPR_Workshops,
author = {Cheng, Guoan and Matsune, Ai and Li, Qiuyu and Zhu, Leilei and Zang, Huaijuan and Zhan, Shu},
title = {Encoder-Decoder Residual Network for Real Super-Resolution},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
} 

Acknowledgements

This code is built on EDSR (PyTorch). We are grateful to the authors for sharing their codes of EDSR.