Home

Awesome

PDASR

Perception-Distortion Balanced ADMM Optimization for Single-Image Super-Resolution
Yuehan Zhang, Bo Ji, Jia Hao, and Angela Yao
In ECCV 2022

Introduction

Single Image Super-Resolution (SISR) usually only does well in either objective quality or perceptual quality, due to the perception-distortion trade-off. In this paper, we proposed a two-stage model trained with low-frequency constraint and designed ADMM algorithm. Experimentally. our method achieve high perfromance in both PSNR/SSIM (objective quality) and NRQM/LPIPS (perceptual quality). Check followings for details.

Paper | Sumpplementary Material

Getting Start

git clone https://github.com/Yuehan717/PDASR  
cd PDASR/src
pip install -r requirements.txt

Data Preparation

Testing

python test.py --scale 4 --save test_results --templateD HAN --templateP Clique \
--dir_data [root of testing sets] --data_test Set5+Set14+B100+Urban100 \
--pre_train ../models/model_trained.pt --save_results

We also provide the testing results in our paper.

Training

Instructions coming soon

Our code is based on EDSR. Thanks to their great work.