Talk:Pixel-art scaling algorithms

License

 * Since Kreed released the source code under the GNU General Public License, it is freely available to anyone wishing to utilize it in an open source project.

I wonder if this is slightly misleading. The GPL doesn't apply to the algorithm itself (that is, you don't have to GPL your source code just because you use 2xSaI). The only thing that could restrict that is a patent, and it's not patented. Though of course you'd have to GPL your source code if you use a significant portion of Kreed's actual code in your own code. Should this be noted in the article, or would this be too obvious to most people to mention? - furrykef (Talk at me) 01:07, 15 February 2006 (UTC)
 * True. I've added the sentence "Developers wishing to use it in a non-open source project would be required to rewrite the algorithm without using any of Kreed's existing code." This should make things more clear. However, I've looked at the code - basically the only thing about GPL is the website which says "2xSaI is free under GPL". It never mentions the license by its full name, or anywhere in the code. So unfortunately for Kreed, it's probably not legally GPL (The GPL actually requires that the full text of the GPL is distributed with the program - this text does not appear anywhere). &mdash;EatMyShortz 11:15, 24 September 2006 (UTC)
 * Actually, the source code comes with the GPL in a file called license.txt. And even if the notice is incorrectly written, the program is still under the GPL, unless it was based on someone else's work before Kreed got to it (in which case it's a copyright violation anyway). A release under a license can be legally valid even if it doesn't comply with the license itself, and only the copyright holder can enforce the terms of a license. Neon  Merlin  02:59, 4 July 2007 (UTC)

More pictures?
If there was ever an article that cried out for pictures, this has to be it - a simple comparison of the various algorithms would be nice, if anyone can think of an easy way to produce one without relying on screenshots of copyrighted software. 81.86.133.45 19:40, 15 July 2007 (UTC)

Someone should create a test case like and run it through every mentioned algorithm for comparison purposes. — Omegatron 18:11, 29 July 2007 (UTC)

I agree. A noob like me doesn't understand anything of what's written in the article, so without pictures this article is useless for people like me... 77.47.102.19 (talk) 01:07, 15 March 2009 (UTC)

I have generated a Scale2x images with a code written by me in VB6. This is the result images :. But I have a question...if the source images is 384x192pixel, why the 3x image have the same resolution?I don't understand... --KymyA (talk) 16:10, 3 June 2009 (UTC)
 * The source image is not 384x192, that is an image magnified 3 times using nearest-neighbour filtering. The original image would be 128x64 pixels. -- Sander (talk) 11:49, 19 March 2010 (UTC)

More explanations
I think the explanations for EPX/Scale/Mame/Eagle are very nice. Could someone please make explanations for 2xSaI superSaI and HQX?

Veritaba (talk) 02:04, 12 March 2009 (UTC)

Is there a scaling algorithm that works well on anti-aliased images? —Preceding unsigned comment added by 71.167.67.139 (talk) 14:48, 15 September 2009 (UTC)

Kopf and Lischinski's spline-based algorithm?
Should there be something about Kopf and Lischinski's spline-based algorithm here? See http://research.microsoft.com/en-us/um/people/kopf/pixelart/index.html -- Resuna (talk) 12:09, 1 June 2011 (UTC)

De-merged from image scaling
I've now extracted this material from the image scaling article, as it had completely overwhelmed the rest of that article, making it impossible to improve it. -- The Anome (talk) 10:15, 3 July 2016 (UTC)

waifu2x
Description is pasted from the waifu2x website (copyvio?). There's a lot more that could be written about this extremely useful piece of software by experts who've had a chance to study it properly. In my experience (yeah, I know it's OR, so I won't add it to the article) the upscaling process works best with clean source images with low grain. Some jpeg artifacts may be retained and upscaled; in some cases these can be reduced post-processing using smoothing effects in conventional image editors, although this will also soften the image somewhat. Although originally designed for 2-dimensional artwork the software also produces very good results with CG art, with some examples retaining good quality after 4x expansion. Although the software cannot of course add detail that wasn't in the original image, it does retain most, if not all, of the original detail at higher scales. Now could somebody confirm all that and write a learned paper online so it can be quoted here?! Lee M (talk) 01:37, 12 August 2016 (UTC)

Are these really that popular / relevant?
I'm wondering, because judging from the images in the article, most of them are really crappy. And nobody I know actually uses the S2xSAI &c. settings in their emulators, most people seem to prefer either no filtering at all or some kind of old CRT filter. — Preceding unsigned comment added by 80.114.146.117 (talk) 01:31, 19 January 2017 (UTC)
 * Well, ScummVM uses hqx by default, which I guess makes it quite popular. — Preceding unsigned comment added by 90.227.25.67 (talk) 11:39, 8 February 2017 (UTC)
 * No it doesn't, I just checked. By default it doesn't do any interpolation at all, and be glad because most pixel artists draw their artwork in such a way that the placement of the pixels suggest detail that isn't actually there, and that tends to get slaughtered by upscaling algorithms, which is a large part of why they look so disappointing. I just installed the most recent version of ScummVM to see if they changed it, but it's still the same as it always was.
 * And what the default is doesn't mean much anyway. If it had been hqx, I think most users would change it immediately. — Preceding unsigned comment added by 77.61.180.106 (talk) 22:44, 16 March 2022 (UTC)

Where's the actual working description of 2xSai, SuperEagle, and Super2xSai?
I can't find it anywhere in this article. There's a working description of the actual algorithm for Advmame2x, Eagle, and for a couple others. However it's missing working descriptions for the algorithms 2xSai, SuperEagle, Super2xSai, and many others. Benhut1 (talk) 21:23, 25 January 2017 (UTC)

Contradiction in the "Super 2×SaI and Super Eagle" section
If you read it, it says both that super 2xSaI blends more than Super Eagle, and also that Super Eagle blends more than Super 2xSaI, about a sentence apart. -- It makes no sense and needs to be fixed. — Preceding unsigned comment added by 50.35.81.237 (talk) 09:14, 13 February 2018 (UTC)

Can I add this image?
I created - can I add it to the page? Also please check it's correct. Thanks. Drummyfish (talk) 17:55, 11 April 2018 (UTC)

Update: no answer so far, so I added the image. If there are objections, remove it and let me know what's wrong. Drummyfish (talk) 13:01, 15 April 2018 (UTC)