keras-team/keras · error · ValueError

The `weights` argument should be either `None` (random initi

Error message

The `weights` argument should be either `None` (random initialization), `imagenet` (pre-training on ImageNet), or the path to the weights file to be loaded.

What it means

With padding=same, reconstruct_patches tolerates output sizes smaller than grid*patch_size (down to just above (grid-1)*p, i.e. at most one patch worth of overlap trimmed), because same padding guarantees coverage. An output_size outside ((g-1)*p, g*p] per axis would leave pixels uncovered or over-cover them, so it is rejected.

Source

Thrown at keras/src/applications/densenet.py:192

        classifier_activation: A `str` or callable.
            The activation function to use
            on the "top" layer. Ignored unless `include_top=True`. Set
            `classifier_activation=None` to return the logits of the "top"
            layer. When loading pretrained weights, `classifier_activation`
            can only be `None` or `"softmax"`.
        name: The name of the model (string).

    Returns:
        A model instance.
    """
    if backend.image_data_format() == "channels_first":
        raise ValueError(
            "DenseNet does not support the `channels_first` image data "
            "format. Switch to `channels_last` by editing your local "
            "config file at ~/.keras/keras.json"
        )
    if not (weights in {"imagenet", None} or file_utils.exists(weights)):
        raise ValueError(
            "The `weights` argument should be either "
            "`None` (random initialization), `imagenet` "
            "(pre-training on ImageNet), "
            "or the path to the weights file to be loaded."
        )

    if weights == "imagenet" and include_top and classes != 1000:
        raise ValueError(
            'If using `weights` as `"imagenet"` with `include_top`'
            " as true, `classes` should be 1000"
        )

    # Determine proper input shape
    input_shape = imagenet_utils.obtain_input_shape(
        input_shape,
        default_size=224,
        min_size=32,
        data_format=backend.image_data_format(),

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Choose output_size per axis in the range ((grid-1)*p, grid*p]; the common correct choice is grid*p or the original pre-pad input size
2. Drop output_size to let Keras infer grid*p
3. If you truly need a smaller output, crop after reconstruction instead of requesting an invalid size

Example fix

before: reconstruct_patches(p, size=(8,8), padding="same", output_size=(24,24)) -> ValueError; after: out = reconstruct_patches(p, size=(8,8), padding="same"); out = out[:, :24, :24, :]

for g, p, o in zip(grid_dims, patch_dims, output_size):
    assert g * p - p < o <= g * p, f"same padding needs o in ({g*p-p}, {g*p}], got {o}"

def same_padding_ok(grid, patch, out) -> bool:
    return all(g * p - p < o <= g * p for g, p, o in zip(grid, patch, out))

When it happens

Trigger: output_size=(24, 24) with grid=4, patch=8 (24 <= (4-1)*8=24, boundary excluded); output_size larger than grid*p (e.g. 40 with grid=4, p=8); axis-wise mistakes where height is fine but width falls below the floor.

Common situations: Reconstructing to the original padded input size after cropping; mixing per-axis conventions (one axis valid, one same); assuming output_size can be arbitrary when padding=same.

Related errors

• Unknown activation function '{activation}' cannot be seriali
• Could not interpret activation function identifier: {identif
• ConvNeXt does not support the `channels_first` image data fo
• If using `weights="imagenet"` with `include_top=True`, `clas
• DenseNet does not support the `channels_first` image data fo

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/6eda9c0d45c5be4f. Report an issue: GitHub.