EDSR (Enhanced Deep Super-Resolution) Matlab port

EDSR (Enhanced Deep Super-Resolution) Single Image Super Resolution Matlab port

EDSR (Enhanced Deep Super-Resolution) Single Image Super Resolution Matlab port.

■ Prerequisites ■

Matlab 2020b
Image Processing toolbox
Statistics and Machine Learning toolbox
Deep Learning Toolbox
Parallel Computing Toolbox

■ How to Test ■

Run EDSR_Test.m that calls EDSR_2xSuperResolution.m

Trained net is loaded on the line 5 of EDSR_2xSuperResolution.m

■ How to Perform EDSR Super-Resolution to your image file ■

Input image MyPicture.jpg should be pristine (not blurred) image. EDSR neural net will upscale the image by 2x.

img = imread("MyPicture.jpg"); % 1024x768 input image
imgSR = EDSR_2xSuperResolution(img);
imwrite(imgSR, "MyPicture_2xEDSR.png"); % 2048x1536 image is outputted

■ How to Train the network using DIV2K dataset ■

Download DIV2K dataset and place it on
DIV2K/DIV2K_train_HR for train data of 800 images
DIV2K/DIV2K_valid_HR for validation data of 100 images

Run CreateTrainingSetAll_DIV2K.m to create DIV2K_MatlabF folder that contains converted mat files.

■ How to Train the network using Flickr2K dataset ■

Run EDSR_Train_DIV2K.m to train and create EDSR_trained_2x_DIV2K.mat
rename it to EDSR_trained_2x.mat

Download Flickr2K dataset and place it on
Flickr2K/Flickr2K_HR for train data of 2650 images

Run CreateTrainingSetAll_Flickr2K.m to create Flickr2K_MatlabF folder that contains converted mat files.

Run EDSR_Train_Flickr2K.m to train and create EDSR_trained_2x_Flickr2K_epoch10.mat
rename it to EDSR_trained_2x.mat

■ Difference from original EDSR ■

1. Processes only Y component.
2. Mean squared error regression.
3. nResBlock==32, nFeatures==128, nResScale==0.05.
4. Only 2x super resolution is implemented.

■ Changelog ■

Version 20210307 1.0.3

・Updated EDSR_trained_2x.mat : EDSR 32 ResBlock Flickr2K dataset trained file.

Version 20210228 1.0.2

・Updated EDSR_trained_2x.mat to Flickr2K dataset trained file to improve subjective SR quality.

Version 20210224 1.0.1

・Fix typo.

Version 20210223 1.0.0

・Initial release.

■ References ■

Enhanced Deep Residual Networks for Single Image Super-Resolution
http://openaccess.thecvf.com/content_cvpr_2017_workshops/w12/papers/Lim_Enhanced_Deep_Residual_CVPR_2017_paper.pdf

Keras implementation of EDSR
https://github.com/hieubkset/Keras-Image-Super-Resolution

Single Image Super-Resolution Using Deep Learning
(VDSR is implemented using Matlab Deep Learning Toolbox)
https://www.mathworks.com/help/images/single-image-super-resolution-using-deep-learning.html