mnist_corrupted

  • Description:

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

Split Examples
'test' 10,000
'train' 60,000
  • Features:
FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
})
@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}
}

mnist_corrupted/identity (default config)

Visualization

mnist_corrupted/shot_noise

Visualization

mnist_corrupted/impulse_noise

  • Config description: Corruption method: impulse_noise

  • Visualization (tfds.show_examples):

Visualization

mnist_corrupted/glass_blur

Visualization

mnist_corrupted/motion_blur

Visualization

mnist_corrupted/shear

Visualization

mnist_corrupted/scale

Visualization

mnist_corrupted/rotate

Visualization

mnist_corrupted/brightness

Visualization

mnist_corrupted/translate

Visualization

mnist_corrupted/stripe

Visualization

mnist_corrupted/fog

Visualization

mnist_corrupted/spatter

Visualization

mnist_corrupted/dotted_line

Visualization

mnist_corrupted/zigzag

Visualization

mnist_corrupted/canny_edges

Visualization