keras-team/keras · error · ValueError

Expected mode to be one of `caffe`, `tf` or `torch`…

Error message

Expected mode to be one of `caffe`, `tf` or `torch`. Received: mode={mode}

What it means

Validation in keras/src/applications/imagenet_utils.preprocess_input: `mode` selects the preprocessing convention (caffe, tf, or torch). Any other string, including None or a misspelling, is rejected because the function cannot know which channel ordering/scaling to apply.

Solutions

1. Guard target_width >= 0 before the call
2. Fix the upstream size computation
3. Handle degenerate small inputs explicitly

Example fix

# before
out = ops.image.crop_images(img, target_width=(w - 2 * c, h - 2 * c))  # w < 2c
# after
tw = w - 2 * c
if tw < 0:
    raise ValueError('width too small')
out = ops.image.crop_images(img, target_width=(tw, h - 2 * c))

if target_width is None or int(target_width) < 0:
    raise ValueError('target_width must be >= 0')

When it happens

Trigger: Passing target_width < 0, typically from a computed subtraction that underflows for small inputs.

Common situations: Dynamic per-sample widths; size-minus-margin arithmetic going negative; copy-pasted shape tuples with a negative entry.

Related errors

• The `weights` argument should be either `None` (random…
• Argument `segment_ids` and `data` should have same leading…
• Argument `segment_ids` should be an 1-D vector, got shape
• Batch dimensions of inputs to `cdist` must be…
• Expected a square matrix. Received non-square input with…

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/21954819882e711d. Report an issue: GitHub.

Appendix: source

Thrown at keras/src/applications/imagenet_utils.py:90

    ValueError: In case of unknown `data_format` argument."""

PREPROCESS_INPUT_RET_DOC_TF = """
      The inputs pixel values are scaled between -1 and 1, sample-wise."""

PREPROCESS_INPUT_RET_DOC_TORCH = """
      The input pixels values are scaled between 0 and 1 and each channel is
      normalized with respect to the ImageNet dataset."""

PREPROCESS_INPUT_RET_DOC_CAFFE = """
      The images are converted from RGB to BGR, then each color channel is
      zero-centered with respect to the ImageNet dataset, without scaling."""


@keras_export("keras.applications.imagenet_utils.preprocess_input")
def preprocess_input(x, data_format=None, mode="caffe"):
    """Preprocesses a tensor or Numpy array encoding a batch of images."""
    if mode not in {"caffe", "tf", "torch"}:
        raise ValueError(
            "Expected mode to be one of `caffe`, `tf` or `torch`. "
            f"Received: mode={mode}"
        )

    if data_format is None:
        data_format = backend.image_data_format()
    elif data_format not in {"channels_first", "channels_last"}:
        raise ValueError(
            "Expected data_format to be one of `channels_first` or "
            f"`channels_last`. Received: data_format={data_format}"
        )

    if isinstance(x, np.ndarray):
        return _preprocess_numpy_input(x, data_format=data_format, mode=mode)
    else:
        return _preprocess_tensor_input(x, data_format=data_format, mode=mode)

View on GitHub (pinned to 7a34a03db6)