Dogway's Filters Packs

[kedautinh12]

I seen DNR is best result just look detail of bath towel in left-top corner

[Blankmedia]

I have some problem with oyster_deringing

I've added float "h" in the function :

oyster_Deringing(clip src, clip ref, int "radius", float "h", float "sigma", float "mse0",

I've change the lesser or equel for greater or equal to 0 here:

Assert(h>=0,                     "Oyster.Deringing: h has to be greater than 0!")

Now "strength2" isn't being defined before being called in line 364.

If I put strength2 = 1.1 before the function call in line 364 I get script error:invalid arguments to function oy_loop.

[Dogway]

Thanks for the report! I will have a look and report back, but Oyster and Plum are both WIP (work in progress) so I'm not positive it works even when fixed. To start with the BM3D plugin isn't fully ported so many settings are missing which may be critical for the correct output, and second I'm not sure mvtools is as robust in AviSynth+ as it is in VapourSynth, one example is 32-bit support which has some issues as reported here.


EDIT: ok updated. Good luck making it work, it hangs on my rig. I guess iterating KNLMeansCL four times isn't actually avisynth friendly.

[tormento]

Originally Posted by Dogway

Check below with the great sample clip from tormento.

A great sample from a great movie.

Anyway, look at the lower white part corner of the top left towel.

Detalis disappear everywhere but in the LFR only.

[Dogway]

Sorry guys, but I don't see what you guys are seeing.

DNR..........................................................................DNR+LFR+DCTR
............

Anyway I have some ideas to improve the DCTFlicker pass.

On another note, yes MatchClip() which is required for prefilter=5 and some other functions as a replacement for Utils-r41's MatchColorFormat() is not yet released (due in TransformsPack). I'm trying to catch up with all the frameprops changes by pinterf.

[tormento]

Originally Posted by Dogway

Sorry guys, but I don't see what you guys are seeing.

This part (present in raw) plain and simply disappears. The only one with some hint of it is DNR+LFR only.

Plus, I am having the error:

Script error: There is no function named 'MatchClip'.
(D:/Programmi/Media/AviSynth+/plugins64/SMDegrain-3.3.5d~Dogway.avsi, line 887)
(D:/Programmi/Media/AviSynth+/plugins64/SMDegrain-3.3.5d~Dogway.avsi, line 217)

Perhaps I have deleted some necessary plugin. Which is which?

[kedautinh12]

I seen only DNR better than DNR+LFR in perfome part of bath towel deatails

[Blankmedia]

My pleasure! I like seafood so it was compelling to use.

[Dogway]

Ok, maybe I did wrong by presenting all the steps, but I like to keep transparent always. My point was DNR+LFR+DCTR improving over DNR/DNR+LFR, which I think it does. There are dozens of different ways to fix the pointed issue, one of them is raising the cutoff but it wasn't my intention to show an example of perfect denoising, because it simply doesn't exist as stated by Didée. I'm trying hard and have some things in mind to improve low frequency denoising.

@tormento, read my previous post. MatchClip is included in yet-to-be-released TransformsPack, which got delayed due to latest avisynth changes in frameprops. Hopefully I can release tomorrow. It replaces MatchColorFormat from Utils-r41 so one less dependency. It's only needed for prefilter=5 though.

[tormento]

Originally Posted by Dogway

read my previous post

Yep I read.

Please, next time, don't implement functions you need to write yet

[Dogway]

It's written but not released. I was holding back all my scripts for over a week solely because MatchClip (read about the conventions disaster here), so I pulled the trigger.

If you are too desperate use this (disclaimer, use latest avs+):

# Matches 'a' to 'b' clip format (size, length, color model, Jab, bitdepth)
function MatchClip (clip a, clip b, string "kernel", string "matrix", bool "length", bool "fast") {

    krn    = Default (kernel, "Spline16")
    fst    = Default (fast,         true)
    mfc    = Default (length,      false) # Match also framecount

    YUY2   = b.IsYUY2()
    rgbb   = b.isRGB()             rgba    = a.isRGB()
    w      = b.width()             wa      = a.width()
    h      = b.height()            ha      = a.height()
    fc     = b.FrameCount()        fca     = a.FrameCount()
    p_type = b.PixelType()         p_typea = a.PixelType()
    bib    = b.BitsPerComponent()  bia     = a.BitsPerComponent()

    isHD    = (w  > 1099 ||  h  > 599)       isHDa   = (wa > 1099 ||  ha > 599)
    isUHD   = (w  > 2599 ||  h  > 1499)      isUHDa  = (wa > 2599 ||  ha > 1499)
    mat     = Default(matrix, rgbb != rgba ? isUHD  ? "Rec2020" : isHD  ? "Rec709" : "Rec601" : Undefined())

        a
        bib > bia ? ConvertBits(bib, fulls=rgba) : last

    if (!fst) {

        ConvertFormat(w, h, fmt_i=p_typea, fmt_o=p_type, cs_in=mat, cs_out="", kernel=krn, gamma=false, tv_range_in=!rgba, tv_range_out=!rgbb, fulls=rgba, fulld=rgbb, show=false)

    } else {

        rgbp   = rgbb && b.isPlanar()
        isScl  = w != wa || h != ha
        mod    = rgba && rgbb || isy(a) && isy(b) ? 1 : 2

        yv411  = b.isYV411()      yv411a = a.isYV411()
        yv444  = b.is444()        yv444a = a.is444()
        fmt    = Format_fuzzy_search(b, p_type,  bib)
        fmta   = Format_fuzzy_search(a, p_typea, bia)
        bc     = bicubic_coeffs(krn)
        krn    = bc[1]>=0. ? "Bicubic" : krn

        cplace  = rgbb || yv444  || yv411  ? "MPEG1" : isUHD  ? "top_left" : "MPEG2"
        cplacea = rgba || yv444a || yv411a ? "MPEG1" : isUHDa ? "top_left" : "MPEG2"
        not42   = LeftStr(fmta[1],2) != "42"

        # Match dimensions
        isScl ? w > h ? RatioResize(w,mode="adjust2w",kernel=krn,mod=mod,b=bc[0],c=bc[1]) : \
                        RatioResize(h,mode="adjust2h",kernel=krn,mod=mod,b=bc[0],c=bc[1]) : last
                        PadResize(w,h,mode="dilate",             mod=mod)

        # ConvertBackToYUY2 bug: https://forum.doom9.org/showthread.php?p=1038027#post1038027
        isy(b) ? rgba ? DotClip([0.298967,0.586421,0.114612]) : ConvertToY(mat)                                                                                 : \
        rgbb   ? rgbp ? Eval(  "ConvertToPlanarRGB            (mat,interlaced=false"+(yv411a ? ""  :  ",ChromaInPlacement=cplacea")+",  chromaresample=krn)")   : \
                        Eval(  "ConvertTo"   +string(fmt[3])+"(mat,interlaced=false"+(yv411a ? ")" :  ",ChromaInPlacement=cplacea,      chromaresample=krn)"))  : \
        yv411         ? Eval(  "ConvertToYV411                (false,mat,"          +(not42 ? ""   : """ChromaInPlacement=cplacea,""")+"chromaresample=krn)")   : \
        YUY2  && rgba ? Eval("""ConvertBackToYUY2             (mat                                                                                       """)   : \
                        Eval(  "ConvertToYUV"+string(fmt[1])+"(false,mat,"          +(not42 ? ""   :   "ChromaInPlacement=cplacea,")+"  chromaresample=krn"+(yv444 ? ")" : ",ChromaOutPlacement=cplace)"))
    }

    bia > bib      ? ConvertBits(bib, dither= 1, fulls=rgbb)   : last
    mfc ? fca > fc ? Trim(last, fc-1) : last+BlankClip(b,fc-1) : last

    # Planar to Interleaved
    YUY2 && !rgba  ? ConvertToYUY2()  : last }

[tormento]

Originally Posted by Dogway

If you are too desperate use this

Not desperate but I can't simply use prefilter=5 with newer features of SMDegrain.

It works, thanks.

[tormento]

Originally Posted by Dogway

use this

Another needed function:

Script error: There is no function named 'Format_fuzzy_search'.
(D:/Programmi/Media/AviSynth+/plugins64/MatchClip~Dogway.avsi, line 35)
(D:/Programmi/Media/AviSynth+/plugins64/MatchClip~Dogway.avsi, line 56)

[Dogway]

That's in TransformsPack. I'm finishing the script (QC) just waiting for pinterf input so I don't do things twice.

[Dogway]

Updated DCTFilter, hope it's fine now:



I will upload tomorrow with TransformsPack, which isn't finished of course but it's fine to play with already.

[tormento]

Originally Posted by Dogway

Updated DCTFilter, hope it's fine now

Much better. Thanks.

Did you alredy implemented atan2 expr from Avisynth+ 3.7.1 test build 27 (20211117) in your scripts? I saw some legacy comments but the functions are still there.

[Dogway]

Yes, I forgot to add in the changelog. So far it's only used for the cylindrical models in TransformsPack. It's a tiny bit faster than my approx. but it's internal and more precise so a winner for me. I don't think 'atan' is yet optimized but I didn't see it used in any of my scripts.

I will keep the legacy functions for reference, this is also some kind of algo spread sheet for learning.

[Dogway]

By the way, I want to introduce SoftLimiter() a filter I have been working on for the last days. It's an inclusive limiter, that means not only it drops out out of range values, but also will include out of range values in a rolloff manner. This is adaptive in the sense that further out of range values will rolloff stronger.
There's also a dynamic mode where you don't need to input the range floor/ceil after inspection. It uses ShowChannels() which works only over 8-bit so precision is not very granular.
dyn=0 is static (default), dyn=1 uses accumulative loose min/max, which is a set and forget safe mode, dyn=2 is more granular and probably better when the source range floor/ceil varies, and dyn=3 which is very granular.



For comparison this is SoftLimiter2() by raffriff42



And here an example with a tailored sample with range floor at 11 and ceil at 245:
Limiter()

SoftLimiter()



I also optimized the HSV, IPT and OkLab conversion functions and they are ready to shine, now work over any bitdepth.
Here's a comparison for the hue perceptual uniformity. Since Luma and Chroma channels are mapped differently I matched those by eye.

HSV

IPT

OkLab

[tormento]

Originally Posted by Dogway

I also optimized the HSV, IPT and OkLab conversion functions and they are ready to shine, now work over any bitdepth.

[VoodooFX]

Originally Posted by Dogway

And here an example with a tailored sample with range floor at 11 and ceil at 245...

Looks nice. What is that Limiter() there? Could you compare SafeColorLimiter too?