Skip to content

mixed_rescale

DiffMode

Bases: IntEnum

Mode used to calculate the difference between rescale and input clip.

Source code in vodesfunc/rescale_ext/mixed_rescale.py 
 8
 9
10
11
12
13
14
15
16
17
18
class DiffMode(IntEnum):
    """Mode used to calculate the difference between rescale and input clip."""

    MAE = 1
    """Mean Absolute Error"""

    MSE = 2
    """Mean Squared Error"""

    GET_NATIVE = 3
    """Weird headcraft from getnative"""

GET_NATIVE = 3 class-attribute instance-attribute

Weird headcraft from getnative

MAE = 1 class-attribute instance-attribute

Mean Absolute Error

MSE = 2 class-attribute instance-attribute

Mean Squared Error

MixedRB

Implementation of MixedRescale for RescaleBuilder(s)

This is just a stop-gap solution until we (mostly Setsu) can cook up something better.

Example Usage:

upscaler = Waifu2x("trt", 1, fp16=True)

builders = [
    RescaleBuilder(src).descale(Bilinear(border_handling=1), 1280, 720),
    RescaleBuilder(src).descale(BicubicSharp, 1280, 720),
]

# This will be run on all of the above
builders = [
    b.double(upscaler)
    .linemask(KirschTCanny, Bilinear, lthr=50 / 255, hthr=150 / 255, inflate_iter=2)
    .errormask(expand=2)
    .downscale(Hermite(linear=True))
    for b in builders
]

mixed = MixedRB(*builders)
rescaled = mixed.get_upscaled()
Source code in vodesfunc/rescale_ext/mixed_rescale.py 
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
class MixedRB:
    """
    Implementation of MixedRescale for RescaleBuilder(s)

    This is just a stop-gap solution until we (mostly Setsu) can cook up something better.

    Example Usage:

    ```py
    upscaler = Waifu2x("trt", 1, fp16=True)

    builders = [
        RescaleBuilder(src).descale(Bilinear(border_handling=1), 1280, 720),
        RescaleBuilder(src).descale(BicubicSharp, 1280, 720),
    ]

    # This will be run on all of the above
    builders = [
        b.double(upscaler)
        .linemask(KirschTCanny, Bilinear, lthr=50 / 255, hthr=150 / 255, inflate_iter=2)
        .errormask(expand=2)
        .downscale(Hermite(linear=True))
        for b in builders
    ]

    mixed = MixedRB(*builders)
    rescaled = mixed.get_upscaled()
    ```
    """

    def __init__(self, *targets: RescBuildMixed, diffmode: DiffMode = DiffMode.GET_NATIVE, crop_diff: bool = True) -> None:
        """
        A naive per-frame diff approach of trying to get the best descale.
        Credits to Setsu for most of this class.
        """
        y = targets[0].funcutil.work_clip

        for i, d in enumerate(targets):
            d.index = i + 1
            d.diffmode = diffmode
            d.crop_diff = crop_diff
            d._add_index_to_clips()

        prop_srcs = [d.get_diff() for d in targets]
        targets_idx = tuple(range(len(targets)))

        blank = core.std.BlankClip(None, 1, 1, vs.GRAY8, y.num_frames, keep=True)

        map_prop_srcs = [blank.std.CopyFrameProps(prop_src).akarin.Expr("x.PlaneStatsAverage", vs.GRAYS) for prop_src in prop_srcs]

        base_frame, idx_frames = blank.get_frame(0), []

        for i in targets_idx:
            fcurr = base_frame.copy()

            fcurr[0][0, 0] = i

            idx_frames.append((i, fcurr))

        def _select(n: int, f: vs.VideoFrame) -> vs.VideoFrame:
            return min(idx_frames, key=lambda i: f[i[0]][0][0, 0])[1]

        _select_clip = blank.std.ModifyFrame(map_prop_srcs, _select)

        def _selector(clips: list[vs.VideoNode | None]) -> vs.VideoNode:
            base = next(filter(None, clips), None)

            if base is None:
                raise ValueError("Requested clip was None")

            base = base.std.BlankClip(keep=True)
            clips = [c or base for c in clips]

            return core.std.FrameEval(base, lambda n, f: clips[f[0][0, 0]], _select_clip)

        self.upscaled = _selector([t.final()[1] for t in targets])
        self.final = self.upscaled

        self.rescaled = _selector([t.rescaled for t in targets])
        self.credit_mask = _selector([t._return_creditmask() for t in targets])
        self.line_mask = _selector([t._return_linemask() for t in targets])

    def get_upscaled(self, *_) -> vs.VideoNode:
        return self.upscaled

__init__(*targets, diffmode=DiffMode.GET_NATIVE, crop_diff=True)

A naive per-frame diff approach of trying to get the best descale. Credits to Setsu for most of this class.

Source code in vodesfunc/rescale_ext/mixed_rescale.py 
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
def __init__(self, *targets: RescBuildMixed, diffmode: DiffMode = DiffMode.GET_NATIVE, crop_diff: bool = True) -> None:
    """
    A naive per-frame diff approach of trying to get the best descale.
    Credits to Setsu for most of this class.
    """
    y = targets[0].funcutil.work_clip

    for i, d in enumerate(targets):
        d.index = i + 1
        d.diffmode = diffmode
        d.crop_diff = crop_diff
        d._add_index_to_clips()

    prop_srcs = [d.get_diff() for d in targets]
    targets_idx = tuple(range(len(targets)))

    blank = core.std.BlankClip(None, 1, 1, vs.GRAY8, y.num_frames, keep=True)

    map_prop_srcs = [blank.std.CopyFrameProps(prop_src).akarin.Expr("x.PlaneStatsAverage", vs.GRAYS) for prop_src in prop_srcs]

    base_frame, idx_frames = blank.get_frame(0), []

    for i in targets_idx:
        fcurr = base_frame.copy()

        fcurr[0][0, 0] = i

        idx_frames.append((i, fcurr))

    def _select(n: int, f: vs.VideoFrame) -> vs.VideoFrame:
        return min(idx_frames, key=lambda i: f[i[0]][0][0, 0])[1]

    _select_clip = blank.std.ModifyFrame(map_prop_srcs, _select)

    def _selector(clips: list[vs.VideoNode | None]) -> vs.VideoNode:
        base = next(filter(None, clips), None)

        if base is None:
            raise ValueError("Requested clip was None")

        base = base.std.BlankClip(keep=True)
        clips = [c or base for c in clips]

        return core.std.FrameEval(base, lambda n, f: clips[f[0][0, 0]], _select_clip)

    self.upscaled = _selector([t.final()[1] for t in targets])
    self.final = self.upscaled

    self.rescaled = _selector([t.rescaled for t in targets])
    self.credit_mask = _selector([t._return_creditmask() for t in targets])
    self.line_mask = _selector([t._return_linemask() for t in targets])