TheAlgorithms/Python · error · TypeError

Invalid image dtype {dtype!r}, expected uint8 or float32

Error message

Invalid image dtype {dtype!r}, expected uint8 or float32

What it means

Raised by the _DataSet constructor when the requested dtype, after being normalized through tf.compat.v1.dtypes.as_dtype, is neither uint8 nor float32. The dataset pipeline only supports these two output dtypes because MNIST pixels are natively uint8 and float32 is the only supported normalized ([0,1]) form. Any other NumPy or TensorFlow dtype is rejected with a TypeError.

Source

Thrown at neural_network/input_data.py:158

        Args:
          images: The images
          labels: The labels
          fake_data: Ignore inages and labels, use fake data.
          one_hot: Bool, return the labels as one hot vectors (if True) or ints (if
            False).
          dtype: Output image dtype. One of [uint8, float32]. `uint8` output has
            range [0,255]. float32 output has range [0,1].
          reshape: Bool. If True returned images are returned flattened to vectors.
          seed: The random seed to use.
        """
        seed1, seed2 = random_seed.get_seed(seed)
        # If op level seed is not set, use whatever graph level seed is returned
        self._rng = np.random.default_rng(seed1 if seed is None else seed2)
        dtype = dtypes.as_dtype(dtype).base_dtype
        if dtype not in (dtypes.uint8, dtypes.float32):
            msg = f"Invalid image dtype {dtype!r}, expected uint8 or float32"
            raise TypeError(msg)
        if fake_data:
            self._num_examples = 10000
            self.one_hot = one_hot
        else:
            assert images.shape[0] == labels.shape[0], (
                f"images.shape: {images.shape} labels.shape: {labels.shape}"
            )
            self._num_examples = images.shape[0]

            # Convert shape from [num examples, rows, columns, depth]
            # to [num examples, rows*columns] (assuming depth == 1)
            if reshape:
                assert images.shape[3] == 1
                images = images.reshape(
                    images.shape[0], images.shape[1] * images.shape[2]
                )
            if dtype == dtypes.float32:
                # Convert from [0, 255] -> [0.0, 1.0].

View on GitHub (pinned to f5988cc097)

Solutions

1. Pass dtype=np.uint8 (pixel values 0-255) or dtype=np.float32 (values 0-1); these are the only supported values
2. If you need another dtype, load with float32 and cast afterwards: images.astype(np.float64)
3. Check for typos in the dtype string (e.g. 'float' instead of 'float32')

Example fix

# before
datasets = read_data_sets('/tmp/mnist', dtype=np.float64)  # TypeError

# after
datasets = read_data_sets('/tmp/mnist', dtype=np.float32)
images64 = datasets.train.images.astype(np.float64)  # cast later if needed

import numpy as np

VALID_DTYPES = (np.uint8, np.float32)

def check_dtype(dtype):
    base = np.dtype(dtype).type
    if base not in VALID_DTYPES:
        raise ValueError(f'{dtype!r} not supported; use uint8 or float32')

def is_supported_dtype(dtype) -> bool:
    return np.dtype(dtype).type in (np.uint8, np.float32)

try:
    datasets = read_data_sets(train_dir, dtype=dtype)
except TypeError as e:
    if 'Invalid image dtype' in str(e):
        dtype = np.float32  # fall back to a supported dtype
        datasets = read_data_sets(train_dir, dtype=dtype)
    else:
        raise

When it happens

Trigger: Calling read_data_sets(..., dtype=...) or constructing _DataSet(..., dtype=...) with values such as np.float64, np.int32, tf.int64, or the string 'float64'. dtypes.as_dtype resolves the name, but the base_dtype falls outside the allowed pair.

Common situations: Porting old tutorials that pass dtype=np.float64 for higher precision, copying a dtype from a different dataset loader (e.g. CIFAR loaders that accept float64), or assuming any NumPy dtype string is accepted.

Related errors

• number must be positive
• The value of input must be non-negative
• Input list must contain at least two elements
• Inputs and select signal must be 0 or 1
• plain must contain only lowercase letters (a-z)

AI-assisted analysis of TheAlgorithms/Python@f5988cc097 (2026-08-14). Data as JSON: /api/errors/2d1405d7f289547a. Report an issue: GitHub.