Mid-February 2020, I got an idea to upscale all images of this game by using a machine learning to improve the quality.
It took a couple of hours to proceeded but the result is pretty nice.
Before vs After
How I validated all images
The game is procedural with a lot of random and we also use post processing (e.g scanline, effects, ...).
So I did a special level (GraphicsDebugger) where I can scroll all objects on screen to see if there are good or not.
Bad conversion and manual retouch
There are some cases where the upscaling with the machine learning didn't work well. It didn't work well when we have background with black and white strip with text.The sprites was better with the new version but I had to use the original version for texts and panel backgrounds.