Log in

View Full Version : GradFun3 vs. f3kdb.Deband


asarian
20th December 2018, 10:08
(Instead of further cluttering up the 'Portable FATPACK' thread with debanding issues, allow me to just put this out here on its own).

I did some more testing on bilateralGPU-assisted GradFun3 vs. f3kdb.Deband. And what can I say? The verdict is, sadly, very much against the former. An image is worth a thousand words:

GradFun3 vs. f3kdb.Deband (https://1drv.ms/u/s!AhSxhQ9g_mrMlDjFQYTuL1x8ZXcg)

The top-image had reasonable GradFun3 settings:

vid = fvs.GradFun3 (src=vid, thr=0.6, radius=16, elast=4.0, smode=5)

As you can tell, the resultant (high-quality) x264 encoding is horrific. Total garbage, really. So, I decided to crank things up a bit:

vid = fvs.GradFun3 (src=vid, thr=7.0, radius=16, elast=4.0, smode=5)

That's a very high 'thr' setting. Still garbage output. And finally, the good old trusted f3kdb.Deband:

vid = core.f3kdb.Deband (vid, preset="medium", dither_algo=2)

As you can immediately tell, f3kdb.Deband is the only one who gets it right.

Now, either I'm doing GradFun3 immensely wrong, or it's simply not practically usable for any of my re-encodings. Please, advise.

Groucho2004
20th December 2018, 11:31
This was subject of a recent discussion in case you missed it:
https://forum.doom9.org/showthread.php?p=1858413#post1858413

asarian
20th December 2018, 11:55
This was subject of a recent discussion in case you missed it:
https://forum.doom9.org/showthread.php?p=1858413#post1858413

Yes, seems I complete missed it. :o Thanks for the link.