Image-enhance / scaleX4 -keras

Image super-resolution (restoration of rich details in a low resolution image)

Setup

Supports Keras with Tensorflow backend.
By default GPU use(recommended)
For CPU only please uncomment in main_dirpath.py 3 rd line
#os.environ['CUDA_VISIBLE_DEVICES'] = '-1'

Model weights http://epsilonsys.com/weights025-17-0.93.rar
Train data https://data.vision.ee.ethz.ch/cvl/DIV2K/

Usage

The model weights can be downloaded from http://epsilonsys.com/weights025-17-0.93.rar to /weights_Double/ folder :
python main_dirpath.py "imgpath", where imgpath is a full path to the images folder.

Our Quantitative results on Set5 with X4 computed on Y from YCbCr

SSIM 0.904
Best SSIM (NTIRE 2017) 0.901

Our Quantitative results on Set5 with X4 computed on RGB

SSIM 0.879
Best SSIM (NTIRE 2017) 0.874

Official SSIM score from NTIRE 2017 :
http://www.vision.ee.ethz.ch/~timofter/publications/Agustsson-CVPRW-2017suppl.pdf