CAS - Contrast Adaptive Sharpening

[HolyWu]

https://github.com/HomeOfVapourSynth...apourSynth-CAS

[ChaosKing]

Is it a new implementation of CAS?

Seems very fast but also happily sharpens artifacts

[HolyWu]

Originally Posted by ChaosKing

Is it a new implementation of CAS?

Seems very fast but also happily sharpens artifacts

I think the implementation is kinda the same as the official one. What's your parameter and the result? It should look ok for small sharpness value.

[ChaosKing]

I used higher values sharpen=0.8 / 1.0
I like the sharpness. Even with high values it produces only minimal oversharpening effects.

CAS(0.8)


Look at the hair ( Don't use zoom in your Browser. Set it to 100% )


p.s. I saw on github that you're adding cas to LSFmod. Will test it later.

[Boulder]

This looks quite interesting indeed Would it be possible to have an Avisynth+ version as well?

[masterkivat]

I really liked this filter, the tests I made gave pretty good results, thank you for this, HolyWu!

[HolyWu]

Update r2:

• Default to process only luma plane for non-RGB formats.
• Change default sharpness to 0.5.

[ChaosKing]

Originally Posted by Boulder

This looks quite interesting indeed Would it be possible to have an Avisynth+ version as well?

Avisynth port is rdy https://github.com/Asd-g/AviSynth-CAS

[Boulder]

Originally Posted by ChaosKing

Avisynth port is rdy https://github.com/Asd-g/AviSynth-CAS

Thanks for the heads up

And also a big thank you to HolyWu for the original version

[feisty2]

you could also make arbitrary filtering contrast adaptive, your own custom sharpening kernels or some other filters, it's basically

# assume clp is an fp32 clip
clpmax = core.std.Maximum(clp)
clpmin = core.std.Minimum(clp)
contrast = core.std.Expr([clpmax, clpmin], "x y - {eps} +".format(eps=1e-8)) # avoid division by zero

flt = core.xxx.Filter(clp) # some filtering here

adaptive_dif = core.std.Expr([flt, clp, contrast], "x y - z / {strength} *".format(strength=0.05))
clp = core.std.MergeDiff(clp, adaptive_dif)

it's not the exact same algorithm but you get the main idea

EDIT:
this one is closer to the original algorithm, its still not the exact same tho.

# assume clp is an fp32 clip
clpmax = core.std.Maximum(clp)
clpmin = core.std.Minimum(clp)
maxneg = core.std.Expr(clpmax, "1 x -")
clpmin = core.std.Expr([clpmin, maxneg], "x y min")
contrast = core.std.Expr([clpmin, clpmax], "x y / 0 max 1 min sqrt")

flt = core.xxx.Filter(clp) # some filtering here

adaptive_dif = core.std.Expr([flt, clp, contrast], "x y - z * {strength} *".format(strength=4.0))
clp = core.std.MergeDiff(clp, adaptive_dif)

[Selur]

btw. just wanted to note that I really like this filter.
Thanks

[sixthofmay]

Can anyone share the binary for Avisynth 2.6 32bit? I prefer 32bit versions (with XP 32bit support) if possible. I searched the web but only found a binary for Vapoursynth (which I've not used before).

If not possible, how do I create a binary? What developent platform was this created on? I'm a developer but mainly Delphi and Lazarus. It looks like MS Visual Studio. How would I begin?

[ChaosKing]

I don't know about xp support but here is an avisynth version https://github.com/Asd-g/AviSynth-CA...ases/tag/1.0.1

[sixthofmay]

Thank you! I installed the 32bit version from the link you posted. It tests good in XP SP3 32bit, Windows7 Pro SP1 64bit, and Windows 10 Pro 64bit. I'm using Avisynth 2.6 32 bit.

I'm checking if CAS can reduce or eliminate the Warpsharping on a video I'm working on.