Blind SR Kernel Estimation with scale factor of 3 using an Internal-GAN

"KernelGAN"

Bojun Ouyang, Junyi Zhou

Paper: https://arxiv.org/abs/1909.06581

Usage:

Generate your own data:

In order to generate the LR images from your own HR images such that its width and height will be 1/3 of the HR images, run the files under the datagenerator folder:

1) put your HR images under the 'datagenerator/input_image' folder, the images must be in jpg format!

2) run the 'generator.py' file, this will generate the LR images by first applying a
   a randomly sample gaussian kernel then downsampling by 3.
 
3) The kenel(ground truth) and the LR images will be under the 'datagenerator/output_kernel' folder

Quick usage on your data:

Put your LR images in the input_image folder, run the train.py file. It will generate the predicted kernel and HR images in the Results folder.