Quoted from Rarehero:
It's not about analyzing every frame. It's about what looks right and what looks wrong. I think if their High res filter was able to interpret things in the way some video game emulator settings do, I might be interested...until then, I'm sticking w/ DotsXL. For instance, this filter on Super Mario World makes the game look like "hand drawn" type art with perfectly curved lines. ColorDMD doesn't do that. I wonder if future updates could be capable of offering a smoother smoothing mode.
This paper was an inspiration for trying to add upscaling to the ColorDMD but the example of SMW graphics is a bit optimistic. The problem with the curve fitting in this algorithm is that it makes everything look curvy... which is kind of cool but it makes text look weird.
The dark little secret of pinball animation is that most of the time you're looking at text. If the text is goofy, people hate it no matter how the graphics look... and one of the main reason people like the smoothing in the ColorDMD is that (most of the time) it makes the text less "pixellated" and more pleasant to look at.
When I first unveiled our upscaler in March of 2013 the feedback I got was the text look blobby. I spent several months tweaking the custom scaling algorithm (it doesn't use curve fitting and does incorporate specific features of the original round dot artwork) just to try to make it look good across text of varying sizes and fonts.
If you're interested in the SMW scaler, here's a reference to the paper and some associated comparisons: https://johanneskopf.de/publications/pixelart/paper/pixel.pdf
It looks great on the curvy bits but really screws up anything that's supposted to have a sharp angle. Also notice that on the turtle shell above, it turnes the lines in the shell into polka dots... looking more like a lady bug.
The challenge is trying to find a tradeoff that handles both text and animation reasonably well while still being implementable with available hadware resources. Newer/larger FPGAs help.