Meta-SR on Keras and Tensorflow 2.0

Introduction

Implementation of (2019) Meta-SR: A Magnification-Arbitrary Network for Super-Resolution.

Nearest	Bicubic	Meta-SR	Real

Attention

• The paper using Matlab bicubic resize for LR image, but here using pillow bicubic resize.
• Pretrained model train on x1.0 to x4.0, if you want test over x4.0 better to retrain the model.

Data

• Download training data from DIV2K
• Get pretrained model from Google Drive

Environment

python==3.6
tensorflow==2.0

How to use

Train

python train.py

Predict

python predict.py --image test.png --model weights.h5 --scale 4.0

To Do

• Add MetaRDN

Reference

• Official implementation in pytorch. link