Near-perfect pixel art upscaling algorithm

If it's not ZDoom, it goes here.

Near-perfect pixel art upscaling algorithm

Postby Darkcrafter » Mon Oct 19, 2020 6:27 am

I once found a demo version of tool that implements an MSU algorithm that now seems to be completely outcasted by modern deep learning solutions like enhance or topaz gigapixel ai. Despite all photorealistic results deep learning algos give they still struggle scaling up pixel art - not just a bunch of collection of fancy pictures but also lot of old games.

The algorithm is called SmartEdge2 and its main quality towards pixel art is just a part of it, the parameter called "Fractal Substitution". I would always stick with bilinear upscaling, giving all the parameters the value of "0" keeping "2" only for fractal substitution.

Of course it's not perfect but the way it restores thing lines and even dithering patterns amaze. Before the upscaling I'd also save the source files palette and then apply it to the result easing the task of further image cleaning.

Under Windows 10 x64 this software needs a compatibity setting of Windows XP SP2 and right of administrator to run. The bitter side here is that the demo version disables the loading of images any higher than 512x512 pixels but if you do the enlargement twice, in series it works.

Here is the link: ... sp=sharing

Mirror #1: ... sp=sharing
Mirror #2: ... sp=sharing

Mirror #3:
Mirror #4:
Mirror #5:

You do not have the required permissions to view the files attached to this post.
User avatar
Joined: 23 Sep 2017
Location: South Russia
Operating System: Windows 10/8.1/8/201x 64-bit
OS Test Version: No (Using Stable Public Version)
Graphics Processor: nVidia with Vulkan support

Return to Off-Topic

Who is online

Users browsing this forum: No registered users and 0 guests