TensorFlow 2.0 RC is available Learn more

mnist_corrupted

MNISTCorrupted is a dataset generated by adding 15 corruptions to the test images in the MNIST dataset. This dataset wraps the static, corrupted MNIST test images uploaded by the original authors

mnist_corrupted is configured with tfds.image.mnist_corrupted.MNISTCorruptedConfig and has the following configurations predefined (defaults to the first one):

  • identity (v0.0.1) (Size: 235.23 MiB): Corruption method: identity

  • shot_noise (v0.0.1) (Size: 235.23 MiB): Corruption method: shot_noise

  • impulse_noise (v0.0.1) (Size: 235.23 MiB): Corruption method: impulse_noise

  • glass_blur (v0.0.1) (Size: 235.23 MiB): Corruption method: glass_blur

  • motion_blur (v0.0.1) (Size: 235.23 MiB): Corruption method: motion_blur

  • shear (v0.0.1) (Size: 235.23 MiB): Corruption method: shear

  • scale (v0.0.1) (Size: 235.23 MiB): Corruption method: scale

  • rotate (v0.0.1) (Size: 235.23 MiB): Corruption method: rotate

  • brightness (v0.0.1) (Size: 235.23 MiB): Corruption method: brightness

  • translate (v0.0.1) (Size: 235.23 MiB): Corruption method: translate

  • stripe (v0.0.1) (Size: 235.23 MiB): Corruption method: stripe

  • fog (v0.0.1) (Size: 235.23 MiB): Corruption method: fog

  • spatter (v0.0.1) (Size: 235.23 MiB): Corruption method: spatter

  • dotted_line (v0.0.1) (Size: 235.23 MiB): Corruption method: dotted_line

  • zigzag (v0.0.1) (Size: 235.23 MiB): Corruption method: zigzag

  • canny_edges (v0.0.1) (Size: 235.23 MiB): Corruption method: canny_edges

mnist_corrupted/identity

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/shot_noise

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/impulse_noise

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/glass_blur

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/motion_blur

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/shear

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/scale

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/rotate

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/brightness

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/translate

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/stripe

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/fog

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/spatter

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/dotted_line

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/zigzag

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

mnist_corrupted/canny_edges

FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})

Statistics

Split Examples
ALL 70,000
TRAIN 60,000
TEST 10,000

Urls

Supervised keys (for as_supervised=True)

(u'image', u'label')

Citation

@article{mu2019mnist,
  title={MNIST-C: A Robustness Benchmark for Computer Vision},
  author={Mu, Norman and Gilmer, Justin},
  journal={arXiv preprint arXiv:1906.02337},
  year={2019}
}