The scribd paper link is broken but you can find it here: http://research.microsoft.com/en-us/um/people/kopf/pixelart/... . It's very interesting and definitely worth a read in my opinion. You've got to hand it to Microsoft, we've seen a lot of great papers come out of their research department in the last 15 years. Here is a video of the algorithm being applied to Super Mario that I found when I was looking for the paper: http://www.youtube.com/watch?v=o2Fd-4NzB0w
I agree that there's nothing wrong with dupes in general (although if they become too prevalent, it can be frustrating), but I think pointing out dupes can be very helpful. If it's a "dupe" of a story that went nowhere, that doesn't provide much information, but if there was a previous discussion, linking to that allows people to find the already existing conversation about it.
The last 2 pages of the paper has 3 comparisons with hq2x. It seems like the new algorithm does appear to be better than hq2x by being smoother and simpler.
DECKARD: Enhance 224 to 176. Enhance, stop. Move in, stop. Pull out, track right, stop. Center in, pull back. Stop. Track 45 right. Stop. Center and stop. Enhance 34 to 36. Pan right and pull back. Stop. Enhance 34 to 46. Pull back. Wait a minute, go right, stop. Enhance 57 to 19. Track 45 left. Stop. Enhance 15 to 23. Give me a hard copy right there.
Though I mostly understand how much of a different algorithm it is, its results are surprisingly similar to hq4x. It sort of looks like a vectorized form - edges are smoothed better, but everything is excessively curvy (seriously, there are almost no corners anywhere). Not quite, as vectorized forms of images tend to mess up horrifically on some corner-meetings (examples of which are in the paper). Which this doesn't seem to do - very impressive!
Does anyone know what hq4x does to the Doom picture? I'd be interested in comparing edge-cases like that, especially as it seems hq4x would handle it decently well, while this algorithm does pretty poorly. Understandably so, but I wonder if it can be adapted to handle broader blending forms.
hq4x is certainly in the sweet-spot for real-time emulation upscaling.
However, this algorithm does actually vectorize the image, which has interesting uses for e.g. running all the sprites of a game through it and using the static results in a remake (they also talk about extending it to fill in animation frames in time which would be extra cool).
i've used potrace many times before - it works really well (i once hand-wrote some "powerpoint" slides, scanned them, then used potrace to make them appear "natural" - people thought i had made a font based on my handwriting!)
maybe it's the handling of colour that's new here? potrace only does monochrome images, as far as i know.
What's new about these, does anyone remember Super Eagle and 2xSAI used in emulators like VBA and ZSNES to stretch out and smoothen up old low pixel graphics at decent fps that gives smooth gameplay?
31 comments
[ 3.6 ms ] story [ 77.9 ms ] thread... If this really works as well as it seems, it's going to ruin some of our best pop-culture-technology jokes.
http://www.youtube.com/watch?v=KUFkb0d1kbU
Does anyone know what hq4x does to the Doom picture? I'd be interested in comparing edge-cases like that, especially as it seems hq4x would handle it decently well, while this algorithm does pretty poorly. Understandably so, but I wonder if it can be adapted to handle broader blending forms.
It doesn't look good, as there's not enough high-contrast edges for hqx to work.
However, this algorithm does actually vectorize the image, which has interesting uses for e.g. running all the sprites of a game through it and using the static results in a remake (they also talk about extending it to fill in animation frames in time which would be extra cool).
hmmm. wikipedia turns up many http://en.wikipedia.org/wiki/Comparison_of_raster_to_vector_... but the one i was thinking of was potrace - http://potrace.sourceforge.net/
i've used potrace many times before - it works really well (i once hand-wrote some "powerpoint" slides, scanned them, then used potrace to make them appear "natural" - people thought i had made a font based on my handwriting!)
maybe it's the handling of colour that's new here? potrace only does monochrome images, as far as i know.
Ref: http://en.wikipedia.org/wiki/Pixel_art_scaling_algorithms