aa

cope_aa(clip, multiplier=None, antialiaser=Eedi3(0.125, 0.25, gamma=65, vthresh0=40, vthresh1=60, field=1, sclip_aa=None), scaler=Lanczos, mode=CopeMode.Inverse, mask=True, no_aa_ranges=[], hybrid_cl=False, **kwargs)

Cope and lazy function to AA a doubled clip. Usually while rescaling. This is probably overall an awful idea.

Parameters:

Name	Type	Description	Default
multiplier	float | None	Basically rfactor. If you're doubling a 720p clip you'll only have a 1440p clip to AA. EEDI3 will fuck it. Defaults to 1.2 if the input clip is smaller than 1700p. 1 otherwise.	None
antialiaser	Antialiaser	Antialiaser used for actually doing the stuff. Defaults to EEDI3 with some somewhat conservative settings kindof.	Eedi3(0.125, 0.25, gamma=65, vthresh0=40, vthresh1=60, field=1, sclip_aa=None)
scaler	KernelT | ScalerT | Sequence[Union[KernelT, ScalerT]]	Scaler(s) or rather kernel(s) in this case. Used to up- and downscale for the rfactor. If you specify a third one it will get used to scale the mask. (Otherwise Bilinear)	Lanczos
mode	CopeMode | int	Method to return back to input res. Available are UpDown (simple downscale), Descale and Inverse (fmtc).	Inverse
mask	bool | VideoNode	Mask for eedi3 possibly save some calculations. Can be a custom one, True for a Kirsch or False to disable.	True
no_aa_ranges	FrameRangesN	Ranges you might not wanna AA for one reason or another.	[]
hybrid_cl	bool	Use eedi3cl on one of the two interpolate calls. Not sure if this is useful or not. Just wrote it to test.	False

Source code in vodesfunc/aa.py

def cope_aa(
    clip: vs.VideoNode,
    multiplier: float | None = None,
    antialiaser: Antialiaser = Eedi3(0.125, 0.25, gamma=65, vthresh0=40, vthresh1=60, field=1, sclip_aa=None),
    scaler: KernelT | ScalerT | Sequence[Union[KernelT, ScalerT]] = Lanczos,
    mode: CopeMode | int = CopeMode.Inverse,
    mask: bool | vs.VideoNode = True,
    no_aa_ranges: FrameRangesN = [],
    hybrid_cl: bool = False,
    **kwargs: KwargsT,
) -> vs.VideoNode:
    """
    Cope and lazy function to AA a doubled clip. Usually while rescaling.
    This is probably overall an awful idea.

    :param multiplier:              Basically rfactor. If you're doubling a 720p clip you'll only have a 1440p clip to AA. EEDI3 will fuck it.
                                    Defaults to 1.2 if the input clip is smaller than 1700p. 1 otherwise.
    :param antialiaser:             Antialiaser used for actually doing the stuff. Defaults to EEDI3 with some somewhat conservative settings kindof.
    :param scaler:                  Scaler(s) or rather kernel(s) in this case. Used to up- and downscale for the rfactor.
                                    If you specify a third one it will get used to scale the mask. (Otherwise Bilinear)
    :param mode:                    Method to return back to input res. Available are UpDown (simple downscale), Descale and Inverse (fmtc).
    :param mask:                    Mask for eedi3 possibly save some calculations. Can be a custom one, True for a Kirsch or False to disable.
    :param no_aa_ranges:            Ranges you might not wanna AA for one reason or another.
    :param hybrid_cl:               Use eedi3cl on one of the two interpolate calls.
                                    Not sure if this is useful or not. Just wrote it to test.
    """

    def fmtc_args(kernel: Kernel) -> KwargsT:
        if isinstance(kernel, Bicubic):
            return KwargsT(kernel="bicubic", a1=kernel.b, a2=kernel.c)
        else:
            return KwargsT(kernel=kernel.__class__.__name__.lower(), taps=kernel.taps if isinstance(kernel, Lanczos) else None)

    if not isinstance(scaler, Sequence):
        scaler = [scaler, scaler]
    scalers = [Kernel.ensure_obj(s) if mode != CopeMode.UpDown else Scaler.ensure_obj(s) for s in scaler]

    if mask is True:
        from vsmasktools import KirschTCanny

        mask = KirschTCanny.edgemask(clip, lthr=60 / 255, hthr=150 / 255, planes=0)

    if not multiplier:
        multiplier = 1.2 if clip.height < 1700 else 1.0
    height = mod2(clip.height * multiplier)
    width = get_w(height, mod=None)
    if mask:
        mask = mask.resize.Bilinear(width, height) if len(scalers) < 3 else scalers[2].scale(mask, width, height)
        mask = mask.std.Binarize(scale_8bit(mask, 16))
    wclip = scalers[0].scale(clip, width, height)
    aa = wclip.std.Transpose()
    aa = antialiaser.interpolate(aa, False, sclip=aa, mclip=mask.std.Transpose() if mask else None, **kwargs).std.Transpose()
    if hybrid_cl and isinstance(antialiaser, Eedi3):
        from copy import deepcopy

        other_antialiaser = deepcopy(antialiaser)
        other_antialiaser.opencl = True
        aa = other_antialiaser.interpolate(aa, False, sclip=aa, **kwargs)
    else:
        aa = antialiaser.interpolate(aa, False, sclip=aa, mclip=mask if mask else None, **kwargs)
    aa = wclip.std.MaskedMerge(aa, mask)
    match mode:
        case CopeMode.Descale:
            aa = scalers[1].descale(aa, clip.width, clip.height)
        case CopeMode.Inverse:
            aa = aa.fmtc.resample(clip.width, clip.height, invks=True, **fmtc_args(scalers[1]))
        case _:
            aa = scalers[1].scale(aa, clip.width, clip.height)
    if not no_aa_ranges:
        return aa
    else:
        try:
            from jvsfunc import rfs

            return rfs(aa, clip, no_aa_ranges)
        except:
            from vstools import replace_ranges as rfs

            return rfs(aa, clip, no_aa_ranges)

pre_aa(clip, radius=1, strength=100, opencl=True, **nnedi3_args)

A prefilter to use in conjunction with an AA function. The idea is to fix the luminance uniformity on lineart and make AAing more effective.

Parameters:

Name	Type	Description	Default
radius	int	Radius used for the unsharp function	1
strength	float	Strength used for the unsharp function	100
opencl	bool	Use nnedi3cl instead of znedi3	True
nnedi3_args		Additional args passed to the respective nnedi function	{}

Returns:

Type	Description
VideoNode	Processed clip

Source code in vodesfunc/aa.py

def pre_aa(clip: vs.VideoNode, radius: int = 1, strength: float = 100, opencl: bool = True, **nnedi3_args) -> vs.VideoNode:
    """
    A prefilter to use in conjunction with an AA function.
    The idea is to fix the luminance uniformity on lineart and make AAing more effective.

    :param radius:      Radius used for the unsharp function
    :param strength:    Strength used for the unsharp function
    :param opencl:      Use nnedi3cl instead of znedi3
    :param nnedi3_args: Additional args passed to the respective nnedi function

    :return:            Processed clip
    """
    args = {"qual": 2, "nsize": 0, "nns": 4, "pscrn": 1}
    args.update(**nnedi3_args)
    clip_y = plane(clip, 0)

    if opencl:
        (left, right, top, bottom) = mod_padding(clip_y, 2, 2)
        width = clip.width + left + right
        height = clip.height + top + bottom
        clip_y = clip_y.resize.Point(width, height, src_left=-left, src_top=-top, src_width=width, src_height=height)

    for i in range(2):
        bob = clip_y.nnedi3cl.NNEDI3CL(field=3, **nnedi3_args) if opencl else clip_y.znedi3.nnedi3(field=3, **nnedi3_args)
        sharp = unsharp_masked(clip_y, radius, strength)
        limit = core.std.Expr([sharp, clip_y, bob[::2], bob[1::2]], "x y z a max max min y z a min min max")
        clip_y = limit.std.Transpose()

    if opencl:
        clip_y = clip_y.std.Crop(left, right, top, bottom)
        clip_y = clip_y.std.CopyFrameProps(clip)

    return clip_y if clip.format.color_family == vs.GRAY else join(clip_y, clip)

CopeMode

Bases: IntEnum

Source code in vodesfunc/aa.py

class CopeMode(IntEnum):
    UpDown = 1
    Descale = 2
    Inverse = 3

Descale = 2 class-attribute instance-attribute

Inverse = 3 class-attribute instance-attribute

UpDown = 1 class-attribute instance-attribute