chipmusic.org is an online community in respect and relation to chip music, art and its parallels.
You are not logged in. Please login or register.
There were things like this before, for example those 2xSAI engines in emulators. They smooth up the pixels and give similar results in real time. Not as smooth, but pretty close.
If this algorithm works in real time it might be good for HD upscaling of old games. Will make them look really cartoony.
ui: These guys are mathematical researchers. Their job is not to showcase pixel artists, but to showcase their algorithm. And for that reason I think it's actually the right choice for them to choose game sprites. The game sprites are neutral. Game sprites should be well polished, but don't really mean anything outside of the game context, which means the average reader will focus on what the algorithm does. More artistic artwork will likely beg the questions, where is this from, what does it mean? and take away the attention of the thing on display (the algorithm.)
And to continue the neutrality argument, game sprites are pixelated because they needed to be, and were actually usually not even perceived as being pixelated because of the TV screen's blur. It was never part of the aesthetic. Pixel art on the other hand, especially in recent years, is focusing on the pixelation as an aesthetic. It might even be seen as a light insult to the artist to depixelate their art. All of a sudden you have two conscious aesthetic choices that are competing, pixelated or depixelated, and the reader will perhaps not be judging the algorithm just on its technical merits because he may have a bias towards the pixelated aesthetic, if they were using "modern" pixel art as examples.
i will quote myself...
...
And yes, they arent pixel artists... but im just saying
Just in case, im not offended or saying this is bad...
...
... but im just saying
hope you get my point
There were things like this before, for example those 2xSAI engines in emulators. They smooth up the pixels and give similar results in real time. Not as smooth, but pretty close.
If this algorithm works in real time it might be good for HD upscaling of old games. Will make them look really cartoony.
I think they wrote that it's too slow to realistically be performed in real time atm. It will enable Nintendo to quickly churn out a bunch of half-assed remakes, for sure, though! 
I wonder if you could write something like this up in CUDA and hardware-accelerate it. I only skimmed the paper so I'm not really sure what the algorithm involves mathematically, but if it were even remotely parallelizable then using hardware acceleration could give them a huge speed boost.
It's not really clear (at least for the purposes of HD upscaling) that you'd even need to do it in real time necessarily. I suppose it depends on how the remake is done: if you re-wrote the engine it would be trivial to do this offline, but if you were emulating it (as I suspect the vast majority of these things would be) it gets a lot harder.
Thanks for posting this, Celsius. Very interesting.
I wonder if this has an application in digital audio aliasing.
And if it were implemented how it would go against current methods.
It would actually be very useful for upscaling small images.
It's more advanced than current algorithms in that it's making important decisions about straight lines, curves, shading differences.If you want to upsize that 200x200 jpg you have, to say 800x800, it's going to look pretty shit, depending on which algorithm you use, (photoshop has bicubic and nearest neighbour, maybe more?) these are somewhat basic ways of upscaling, by either creating a square or many pixels to represent the one, or by shifting colours of pixels between the two original pixels.
This method seems to have some intelligence in deciding where to put a straight line, where things should be shaded or just a sharp colour change.
To apply it specifically to video game sprites seems ridiculous. Video game sprites aren't downscaled, they're hand drawn specifically to fit the color requirements of the console or computer. If you were to test an algorithm for upscaling, it would make much more sense to actually work with downsized, pixelated photos, where the technology would be practically applied.
Try reading the "previous work" section. It lists previous research and applications of upscaling and vectorizing algorithms, fields where the work is highly relevant. You should also try reading the conclusions. Please, at least skim over the whole paper before assuming that the research is useless.
True, rather ignorant of me to do so, but in my defense, 3:30 am is not the my optimal time for reading research papers, let alone expressing rational opinions about them. The one useful application I could find was porting video games to other systems, but even that reason still makes little sense to me, as most video games currently utilize 3d graphics. Even when not using 3d graphics, it's probably more consistent to hand draw new sprites instead of relying on an algorithm that will always have some percentage of failure. Brand new art would also gives the creators a chance to renovate the game's design and keep interest in the franchise.
EDIT: Though, yes, it did pop into my mind that an HD cartoon-y remake of Super Mario World and other classics would be pretty neat. I just think developing an algorithm to do all the work seems like the wrong way to go about it.
Last edited by Andrew Winzenburg (May 27, 2011 6:15 am)
next up:
Converter for NSF songs to make them sound like they were played by real orchestra.
To apply it specifically to video game sprites seems ridiculous. Video game sprites aren't downscaled, they're hand drawn specifically to fit the color requirements of the console or computer.
Ridiculous or not, there's a whole bunch of emulators that already do upscaling (hqx or eagle algorithms for example), and a lot of users enjoying it. Whether you (or the original artists) like what it does to the original artwork or not, there's obviously a demand for effortless upscaling.
If you were to test an algorithm for upscaling, it would make much more sense to actually work with downsized, pixelated photos, where the technology would be practically applied.
There is a whole lot of effort put into upscaling photographic images (in terms of research and programs), but this is not one of them. As explained in the paper, this algorithm is not particularly useful for (relatively) low-contrast images (like the doom face example). It's designed for the no-nonsense kind of pixel art you find in old video games (read 4.1). It's a very specific algorithm for a quite narrow field of interest.
The one useful application I could find was porting video games to other systems, but even that reason still makes little sense to me, as most video games currently utilize 3d graphics.
Even if most current games use 3D graphics (a dubious claim, IMO, especially since you're talking about ports of old video games), the (potentially few) applications it DOES have aren't affected.
Last edited by boomlinde (May 27, 2011 9:27 am)
next up:
Converter for NSF songs to make them sound like they were played by real orchestra.
http://www.zophar.net/utilities/convert … 2midi.html
KeFF wrote:next up:
Converter for NSF songs to make them sound like they were played by real orchestra.
http://www.zophar.net/utilities/convert … 2midi.html
Midi is not orchestra 
Bear in mind though, this whole thing is less "Check out this technology that we created specifically for videogames" and more "Check out how fuckin awesome we are at math and stuff".
I'm pretty sure the choice of game sprites was arbitrary at best, or maybe because they are simple enough to prove the algorithm with little error.
[edit] The choice to use game sprites as an example though will probably catch the eye of a lot of developers who have been trying to to the same thing (and kind of sucking at it) for years now, though.
This is basically one step closer to that silly "ENHANCE" button in every cop procedural coming true.
Last edited by Dauragon (May 27, 2011 1:15 pm)
next up:
Converter for NSF songs to make them sound like they were played by real orchestra.
that's called virt