The Master property contains the indications to create a so called “master image” based on the original image from which the optimization happens.

In this step, a low resolution image can be up-scaled to 4x it’s original size. The algorithm for this uses machine learning, and produces very sharp contours.

- srgan-dt-x4: Use classic SRGAN, trained on the DIVK image
dataset, with 4x upscaling.
After that, a color correction step and a noise removal steps
are applied to the up-scaled image.
Use this value of the enumeration if you want to produce
up-scaled versions of images starting from a lower resolution
image which is the best you have available.
- identity: Do not change the image, use the original value.
This is the algorithm to use if you are starting from a
high-resolution image and want to shrink it down to some
manegeable size.


Updated: