keras-team/keras · error · NotImplementedError

Unknown `fill_mode` {fill_mode}. Expected of one {self._SUPP

Error message

Unknown `fill_mode` {fill_mode}. Expected of one {self._SUPPORTED_FILL_MODES}.

What it means

RandomElasticTransform validates fill_mode in __init__ against _SUPPORTED_FILL_MODES. Fill mode controls how pixels mapped outside the input boundary are filled after the elastic warp; only the enumerated modes (e.g. 'reflect', 'wrap', 'constant', 'nearest') are implemented. Any other string raises NotImplementedError at construction.

Source

Thrown at keras/src/layers/preprocessing/image_preprocessing/random_elastic_transform.py:113

    ):
        super().__init__(data_format=data_format, **kwargs)
        self._set_factor(factor)
        self.scale = self._set_factor_by_name(scale, "scale")
        self.interpolation = interpolation
        self.fill_mode = fill_mode
        self.fill_value = fill_value
        self.value_range = value_range
        self.seed = seed
        self.generator = SeedGenerator(seed)

        if interpolation not in self._SUPPORTED_INTERPOLATION:
            raise NotImplementedError(
                f"Unknown `interpolation` {interpolation}. Expected of one "
                f"{self._SUPPORTED_INTERPOLATION}."
            )

        if fill_mode not in self._SUPPORTED_FILL_MODES:
            raise NotImplementedError(
                f"Unknown `fill_mode` {fill_mode}. Expected of one "
                f"{self._SUPPORTED_FILL_MODES}."
            )

        if self.data_format == "channels_first":
            self.height_axis = -2
            self.width_axis = -1
            self.channel_axis = -3
        else:
            self.height_axis = -3
            self.width_axis = -2
            self.channel_axis = -1

    def _set_factor_by_name(self, factor, name):
        error_msg = (
            f"The `{name}` argument should be a number "
            "(or a list of two numbers) "
            "in the range "

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Use an exact string from RandomElasticTransform._SUPPORTED_FILL_MODES, e.g. 'constant' when pairing it with fill_value
2. Check the accepted set: print(layers.RandomElasticTransform._SUPPORTED_FILL_MODES)
3. Map your existing config vocabulary: OpenCV 'replicate' -> 'nearest', 'edge' -> 'nearest'

Example fix

# before
layers.RandomElasticTransform(fill_mode="replicate")
# after
layers.RandomElasticTransform(fill_mode="nearest")

from keras.src.layers.preprocessing.image_preprocessing.random_elastic_transform import RandomElasticTransform
assert fill_mode in RandomElasticTransform._SUPPORTED_FILL_MODES

def is_valid_fill_mode(v: str) -> bool:
    return v in {"reflect", "wrap", "constant", "nearest"}

try:
    layer = RandomElasticTransform(fill_mode=mode)
except NotImplementedError as e:
    raise ValueError(f"bad fill_mode {mode!r}: {e}") from e

When it happens

Trigger: layers.RandomElasticTransform(fill_mode='replicate') or fill_mode='edge' (Pillow/OpenCV vocabulary), fill_mode='black', or a typo like 'constnat'.

Common situations: Porting augmentation configs from albumentations, torchvision, or OpenCV, whose fill-mode names ('replicate', 'edge') differ from Keras's; assuming OpenCV's cv2.BORDER_* constants work unchanged.

Related errors

• Unknown `interpolation` {interpolation}. Expected of one {se
• The `{name}` argument should be a number (or a list of two n
• The `{name}` argument should be a number (or a list of two n
• The `fill_value` argument should be a number (or a list of t
• The `{name}` argument should be a number (or a list of two n

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/22e03610cf363b87. Report an issue: GitHub.