Posts: 10.549
Threads: 57
Joined: May 2017
When downscaling by a larger factor DPIR might be interessting.
Also applying CAS 0.5-0.7 or DetailSharpen might also help to improve things.
Cu Seiur
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
Posts: 24
Threads: 3
Joined: Jul 2021
(08.08.2021, 18:38)Selur Wrote: When downscaling by a larger factor DPIR might be interessting.
Also applying CAS 0.5-0.7 or DetailSharpen might also help to improve things.
Cu Seiur
Sorry for the very late reply. I've been experimenting a lot with the GLSL resizer I mentioned, and I checked out CAS and DPIR again.
CAS of .5 was better than I would expect for sharpening, and DPIR with a Lambda of 1.33 was also pretty good.
the GLSL SSIMD took a fair amount of messing around with, but setting dscale to nearest neighbor of all things manages to really pull out a ton of detail. Unfortunately when blown up ad compared at the same zoom as the original 4k, it just doesn't have the same visual appearance as the source, so I'm still not quite replicating MadVR's SSIMD. The author's apparently tuned it for Mitchell which is understandable, but for now I've decided to stick with the SSIMD you kindly added in earlier. It emphasizes structural features just enough without ringing, aliasing, or anything like that.