Now VDP limitations are taken into account, but the k-means and the VDP "quantization" are in cascade
I need to speed up a lot the VDP "quantization" in order to include it in the k-means loop
only in this way we can really improve WRT the standard bmp->scr2 converters
BTW, how does the preliminary result look like?
IMO it looks non excellent but convincing ...
The right side should be pure MSX1 screen2 !!
(at least I did my best to include in my code all the limits of MSX1 - tell me if the color clash and the palette are wrong)