denoise

VMDegrain(src, thSAD=60, prefilter=2, smooth=True, block_size=32, overlap=16, **kwargs)

Just some convenience function for mvtools with a useable preset and temporal smoothing. Check the MVTools Docs for the params that aren't listed below.

Parameters:

Name	Type	Description	Default
src	VideoNode	Input to denoise	required
smooth	bool	Run TTempsmooth on the denoised clip if True	True

Returns:

Type	Description
VideoNode	Denoised clip

Source code in vodesfunc/denoise.py

def VMDegrain(
    src: vs.VideoNode,
    thSAD: int = 60,
    prefilter: vs.VideoNode | int = 2,
    smooth: bool = True,
    block_size: int = 32,
    overlap: int = 16,
    **kwargs: KwargsT,
) -> vs.VideoNode:
    """
    Just some convenience function for mvtools with a useable preset and temporal smoothing.
    Check the MVTools Docs for the params that aren't listed below.


    :param src:             Input to denoise
    :param smooth:          Run TTempsmooth on the denoised clip if True
    :return:                Denoised clip
    """
    from vsdenoise import MVTools, SADMode, SearchMode, MotionMode, PelType, Prefilter

    if isinstance(prefilter, int):
        prefilter = Prefilter(prefilter)
    y = depth(get_y(src), 16)
    d_args = KwargsT(
        prefilter=prefilter,
        thSAD=thSAD,
        block_size=block_size,
        overlap=overlap,
        sad_mode=SADMode.SPATIAL.same_recalc,
        search=SearchMode.DIAMOND,
        motion=MotionMode.HIGH_SAD,
        pel_type=PelType.BICUBIC,
        rfilter=2,
        sharp=2,
    )
    d_args.update(**kwargs)

    out = MVTools.denoise(y, **d_args)
    if smooth:
        out = out.ttmpsm.TTempSmooth(maxr=1, thresh=1, mdiff=0, strength=1)

    out = depth(out, get_depth(src))
    return out if src.format.color_family == vs.GRAY else join(out, src)  # type: ignore

schizo_denoise(src, sigma=[0.8, 0.3], thSAD=60, radius=2, nlm_a=2, prefilter=2, cuda=True, csharp=False, **kwargs)

Convenience function for (k)nlm on chroma and mvtools + bm3d(cuda) on luma. Mostly for personal scripts so please don't complain too much unless it's an actual issue.

Parameters:

Name	Type	Description	Default
src	VideoNode	Input to denoise	required
sigma	float | list[float]	Essentially strength for NLMeans and BM3D. Float or list of floats in this order [bm3d, nlm_uv] or [bm3d, nlm_u, nlm_v]	[0.8, 0.3]
thSAD	int	Not exactly strength but something like that, for mvtools.	60
radius	int | list[int]	Temporal Radius used for NLMeans and BM3D. Int or list of ints in this order [bm3d, nlm]	2
prefilter	VideoNode | int	vsdenoise Prefilter or prefiltered clip to use for mvtools. Defaults to MINBLUR3	2
cuda	bool | list[bool]	Uses NlmCuda and BM3DCuda respectively if available. The latter prefers RTC if available. Will fallback to BM3DHip if installed and no cuda available.	True
csharp	int | bool	Apply contrasharpening after denoising. True defaults to 3 while False obviously disables it.	False
kwargs		Any parameters you might wanna pass to bm3d or mvtools.	{}

Returns:

Type	Description
VideoNode	Denoised clip

Source code in vodesfunc/denoise.py

def schizo_denoise(
    src: vs.VideoNode,
    sigma: float | list[float] = [0.8, 0.3],
    thSAD: int = 60,
    radius: int | list[int] = 2,
    nlm_a: int = 2,
    prefilter: vs.VideoNode | int = 2,
    cuda: bool | list[bool] = True,
    csharp: int | bool = False,
    **kwargs,
) -> vs.VideoNode:
    """
    Convenience function for (k)nlm on chroma and mvtools + bm3d(cuda) on luma.
    Mostly for personal scripts so please don't complain too much unless it's an actual issue.

    :param src:         Input to denoise
    :param sigma:       Essentially strength for NLMeans and BM3D.
                        Float or list of floats in this order [bm3d, nlm_uv] or [bm3d, nlm_u, nlm_v]
    :param thSAD:       Not exactly strength but something like that, for mvtools.
    :param radius:      Temporal Radius used for NLMeans and BM3D.
                        Int or list of ints in this order [bm3d, nlm]
    :param prefilter:   vsdenoise Prefilter or prefiltered clip to use for mvtools.
                        Defaults to MINBLUR3
    :param cuda:        Uses NlmCuda and BM3DCuda respectively if available. The latter prefers RTC if available.
                        Will fallback to BM3DHip if installed and no cuda available.
    :param csharp:      Apply contrasharpening after denoising. True defaults to 3 while False obviously disables it.
    :param kwargs:      Any parameters you might wanna pass to bm3d or mvtools.

    :return:            Denoised clip
    """
    if src.format.color_family != vs.YUV:  # type: ignore
        raise ValueError("schizo_denoise: This function expects a full YUV clip.")

    if not isinstance(radius, list):
        radius = [radius, radius]

    if not isinstance(sigma, list):
        sigma = [sigma, sigma]

    if not isinstance(cuda, list):
        cuda = [cuda, cuda]

    if isinstance(prefilter, int):
        from vsdenoise import Prefilter

        prefilter = Prefilter(prefilter)

    clip = depth(src, 16)

    nlmfunc = core.knlm.KNLMeansCL if not hasattr(core, "nlm_cuda") or not cuda[0] else core.nlm_cuda.NLMeans

    if len(sigma) == 3:
        clip_u = nlmfunc(clip, a=nlm_a, d=radius[1], h=sigma[1], channels="U")
        clip_v = nlmfunc(clip, a=nlm_a, d=radius[1], h=sigma[2], channels="V")
        nlm = join(get_y(clip), get_u(clip_u), get_v(clip_v))  # type: ignore
    else:
        clip_uv = nlmfunc(clip, a=nlm_a, d=radius[1], h=sigma[1], channels="UV")
        nlm = join(clip, clip_uv)  # type: ignore

    # 'Extract' possible bm3d args before passing kwargs to mvtools :)
    bm3dargs = dict(
        block_step=kwargs.pop("block_step", 8),
        bm_range=kwargs.pop("bm_range", 9),
        ps_num=kwargs.pop("ps_num", 2),
        ps_range=kwargs.pop("ps_range", 4),
        fast=kwargs.pop("fast", True),
    )

    y = get_y(clip)
    mv = VMDegrain(y, thSAD, prefilter, **kwargs)

    has_cuda = hasattr(core, "bm3dcuda") or hasattr(core, "bm3dcuda_rtc")
    has_hip = hasattr(core, "bm3dhip")

    if cuda[1] and (has_cuda or has_hip):
        if has_cuda:
            bm3dfunc = core.bm3dcuda if not hasattr(core, "bm3dcuda_rtc") else core.bm3dcuda_rtc
        else:
            bm3dfunc = core.bm3dhip
    else:
        bm3dargs.pop("fast")
        bm3dfunc = core.bm3dcpu

    bm3d = bm3dfunc.BM3Dv2(depth(y, 32), depth(mv, 32), sigma[0], radius=radius[0], **bm3dargs)

    out = join(depth(bm3d, 16), nlm)  # type: ignore
    out = depth(out, get_depth(src))
    if csharp != False:
        out = contrasharpening(out, src, mode=3 if csharp == True else csharp)
    return out