kodemum
commented
6 months ago Yeah, it is definitely far from perfect. The main issue causing what you describe, I think, is that the Geometrize algorithm doesn't have a concept of layers.
What it does is basically, for each symbol, look for a symbol that causes the overall image to move the most towards the target image. This means that if your image is just one red circle with a blue one inside, a circle with a color that is in between will overall move it more towards the target that what we as people would intuitively do, which is to make the first circle fully red and the second one blue.
I did attempt to change it so that instead of using the average color of the area a symbol would cover, it would use the color that appears the most - but this wasn't consistently better either.
If you have some suggestion as to how it could be improved, feel free to let me know!
The conversion works, but most of the times it doesn't, mainly because it uses too many symbols for things that could have been only 1, like instead of using one circle for the eye it ends up using 5 symbols. Some photo memes ends up looking clearer than black n white 2d.