Posts: 6
Threads: 1
Joined: Nov 2023
14.11.2023, 13:42
I'd like to change the sharpening value of CAS from 0.6 to 0.4 in the 2 Step fmtconv resizer from Vapoursynth tab
Hybrid generated vapoursynth configuration :
# Step 1: current: 3840x1608, target: 2324x1212
clip = core.fmtc.resample(clip=clip, kernel="spline36", w=2324, h=1212, interlaced=False, interlacedd=False)
# sharpening stepped resize using CAS
# contrast sharpening using CAS
clip = core.cas.CAS(clip=clip, sharpness=0.6)
# Step 2: current: 2324x1212, target: 1920x804
clip = core.fmtc.resample(clip=clip, kernel="spline36", w=1920, h=804, interlaced=False, interlacedd=False)
Thanks !Please, read the 'Infos needed to fix&reproduce bugs,..'-sticky before you post about a problem.
Posts: 10.522
Threads: 56
Joined: May 2017
This atm. is not supported in Hybrid.
I'll look into how complicated / how much code changes would be needed to change this.
Cu Selur
Posts: 6
Threads: 1
Joined: Nov 2023
Thanks ! It would be a useful feature indeed.... 0.6 is selected because results look closer to the source ?
Posts: 10.522
Threads: 56
Joined: May 2017
No, 0.6 is selected since that is usually my start value if I use CAS.
Cu Selur
Posts: 10.522
Threads: 56
Joined: May 2017
Send you a link, via pm, to a dev version which adds "CAS(sharpness=X)" and "aWarpSharp2(depth=X) as options.
Cu Selur
Posts: 6
Threads: 1
Joined: Nov 2023
So nice of you to include the options in developer build . Will try this one asap.
Downscaling from a grainy 4k to 1080p using spline introduced obvious loss in sharpness a bit. I tried compensating it with CAS=0.6. The results were good but also over sharpened when compared to the source. Single step downscaling with CAS as a filter did not produce the desired results.
This developer build will definitely help. Thanks !
Posts: 10.522
Threads: 56
Joined: May 2017
You might also want to try DPID as resizer when downscaling.
Cu Selur
Posts: 6
Threads: 1
Joined: Nov 2023
Thanks for the developer build. it works as intended. SAR is now set by default as 1:1. That was the only other difference i could find from previous builds.
Thanks for suggesting DPID downscaler. did some tests and it seems to pull more details than spline, even at default lambda 1.0
I think now it just comes down to personal preference between Spline+CAS and DPID. But DPID easily beats the Spline + CAS in transparency
Posts: 10.522
Threads: 56
Joined: May 2017
Happy that helped.
Cu Selur
Posts: 6
Threads: 1
Joined: Nov 2023
Have been playing with DPID for a while. It is the best downscaler so far to reproduce the original sharpness. However, i am seeing vertical line like blocking artifacts (like threads) in grainy content. any x265 tips to avoid it while keeping deblock at -3,-3 ?