xBRZ scaling is pretty darn good. Maybe we can get David to add this ability in G'MIC.


http://sourceforge.net/projects/xbrz/

Would the GMIC filter xBR2x by Garagecoder be w2hat you are looking for?

Ah; had no idea it was there. Need to add a 2x,3x,4x,5x slider that this program has and do a better job addressing the stairstep, but Garagecoder's implemention is nice. Thanks for pointing that out bilbo9955.

Yes; I did. Still would be cool to have the slider Graechan. Also, note the stair-stepping with the G'MIC implementation. Not sure why that is since the stand-alone program is pretty clean.

Lyle, xBRZ seems excellent (maybe GC filter also, not compared yet)
but, as far as xBRZ is concerned, during my initial test I didn't find a way to setup output name to be the same as input name, I have always to copy the name before executing xBRZ then replacing with a paste the "output" name
Did you find some parameter default setting to permanently change this behaviour?

edit: made a comparison with GC filter (run twice): seem to be perfectly equivalent!

Again, other then the stair-stepping issue, the results pretty good with the G'MIC implementation Dinasset.

Actually the way I created gmic xbr2x was a bit of a nightmare (started so had to finish!). As it says on the filter, everything is almost a direct translation of C# source from "imageresizer" (http://code.google.com/p/2dimagefilter/ ... ageResizer) and I happen to not know C#. Also it doesn't include the "blend" option, which might have something to do with the difference in output.

I'm certain it and the other multipliers could be done much more native to gmic, but the algorithm would need a bit of a rethink since it's all based around iterating over a "kernel" which isn't the most efficient way in gmic a lot of the time - I just wanted the output to be correct.

Anyway, yes they could be done but definitely don't count on me to do it!

Thanks for posting GC; other then no multiplier argument, the only thing I see different between the stand-alone and your preset is the stairstepping. Almost straight with the stand-alone, but clearly visible with the G'MIC preset. Don't know if there's some additional processing going on with the stand-alone to straighten the angled lines. Also, with Graechan's second example, I see a clear halo; not sure that is caused by any additional processing that Graechan did. I went ahead and ran the stand-alone twice (x5 then x2) to share that the lines have little aliasing (stair-step) issues).

Dang, I hate it when something isn't quite right

I might end up having a look but the thought of working it out again makes me shudder, maybe a beer or two will help. Any seasoned gmic filter writer who looks at the code is probably going to run for the nearest tree line!

lol; no reason to fret GC. Your implementation still gives better results the scale2x algorithm imo. Again, not sure what the stand-alone program does to address the angled line stair-stepping issue. It may not actually be part of the conversion method itself.

Where do we send the beer to?