Corrupted MNIST dataset.

PiperOrigin-RevId: 257034402

Author: normster

tensorflow_datasets/image/__init__.py

@@ -47,6 +47,7 @@
 from tensorflow_datasets.image.mnist import FashionMNIST
 from tensorflow_datasets.image.mnist import KMNIST
 from tensorflow_datasets.image.mnist import MNIST
+from tensorflow_datasets.image.mnist_corrupted import MNISTCorrupted
 from tensorflow_datasets.image.omniglot import Omniglot
 from tensorflow_datasets.image.open_images import OpenImagesV4
 from tensorflow_datasets.image.oxford_flowers102 import OxfordFlowers102

tensorflow_datasets/image/mnist.py

@@ -35,7 +35,8 @@
 _MNIST_TEST_DATA_FILENAME = "t10k-images-idx3-ubyte.gz"
 _MNIST_TEST_LABELS_FILENAME = "t10k-labels-idx1-ubyte.gz"
 _MNIST_IMAGE_SIZE = 28
-_MNIST_IMAGE_SHAPE = (_MNIST_IMAGE_SIZE, _MNIST_IMAGE_SIZE, 1)
+MNIST_IMAGE_SHAPE = (_MNIST_IMAGE_SIZE, _MNIST_IMAGE_SIZE, 1)
+MNIST_NUM_CLASSES = 10
 _TRAIN_EXAMPLES = 60000
 _TEST_EXAMPLES = 10000
 
@@ -107,8 +108,8 @@ def _info(self):
         builder=self,
         description=("The MNIST database of handwritten digits."),
         features=tfds.features.FeaturesDict({
-            "image": tfds.features.Image(shape=_MNIST_IMAGE_SHAPE),
-            "label": tfds.features.ClassLabel(num_classes=10),
+            "image": tfds.features.Image(shape=MNIST_IMAGE_SHAPE),
+            "label": tfds.features.ClassLabel(num_classes=MNIST_NUM_CLASSES),
         }),
         supervised_keys=("image", "label"),
         urls=[self.URL],
@@ -188,7 +189,7 @@ def _info(self):
                      "classes."),
         features=tfds.features.FeaturesDict({
             "image":
-                tfds.features.Image(shape=_MNIST_IMAGE_SHAPE),
+                tfds.features.Image(shape=MNIST_IMAGE_SHAPE),
             "label":
                 tfds.features.ClassLabel(names=[
                     "T-shirt/top", "Trouser", "Pullover", "Dress", "Coat",
@@ -217,7 +218,7 @@ def _info(self):
                      "when creating Kuzushiji-MNIST."),
         features=tfds.features.FeaturesDict({
             "image":
-                tfds.features.Image(shape=_MNIST_IMAGE_SHAPE),
+                tfds.features.Image(shape=MNIST_IMAGE_SHAPE),
             "label":
                 tfds.features.ClassLabel(names=[
                     "o", "ki", "su", "tsu", "na", "ha", "ma", "ya", "re", "wo"
@@ -319,7 +320,7 @@ def _info(self):
             "matches the MNIST dataset."),
         features=tfds.features.FeaturesDict({
             "image":
-                tfds.features.Image(shape=_MNIST_IMAGE_SHAPE),
+                tfds.features.Image(shape=MNIST_IMAGE_SHAPE),
             "label":
                 tfds.features.ClassLabel(
                     num_classes=self.builder_config.class_number),

tensorflow_datasets/image/mnist_corrupted.py

@@ -0,0 +1,193 @@
+# coding=utf-8
+# Copyright 2019 The TensorFlow Datasets Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Corrupted MNIST Dataset.
+
+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.
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+
+import numpy as np
+import tensorflow as tf
+from tensorflow_datasets.core import api_utils
+from tensorflow_datasets.image import mnist
+import tensorflow_datasets.public_api as tfds
+
+_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
+"""
+
+_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}
+}
+"""
+
+_DOWNLOAD_URL = 'https://zenodo.org/record/3239543/files/mnist_c.zip'
+_CORRUPTIONS = [
+    'identity',
+    'shot_noise',
+    'impulse_noise',
+    'glass_blur',
+    'motion_blur',
+    'shear',
+    'scale',
+    'rotate',
+    'brightness',
+    'translate',
+    'stripe',
+    'fog',
+    'spatter',
+    'dotted_line',
+    'zigzag',
+    'canny_edges',
+]
+_DIRNAME = 'mnist_c'
+_TRAIN_IMAGES_FILENAME = 'train_images.npy'
+_TEST_IMAGES_FILENAME = 'test_images.npy'
+_TRAIN_LABELS_FILENAME = 'train_labels.npy'
+_TEST_LABELS_FILENAME = 'test_labels.npy'
+
+
+class MNISTCorruptedConfig(tfds.core.BuilderConfig):
+  """BuilderConfig for MNISTcorrupted."""
+
+  @api_utils.disallow_positional_args
+  def __init__(self, corruption_type, **kwargs):
+    """Constructor.
+
+    Args:
+      corruption_type: string, name of corruption from _CORRUPTIONS.
+      **kwargs: keyword arguments forwarded to super.
+    """
+    super(MNISTCorruptedConfig, self).__init__(**kwargs)
+    self.corruption = corruption_type
+
+
+def _make_builder_configs():
+  """Construct a list of BuilderConfigs.
+
+  Construct a list of 15 MNISTCorruptedConfig objects, corresponding to
+  the 15 corruption types.
+
+  Returns:
+    A list of 15 MNISTCorruptedConfig objects.
+  """
+  config_list = []
+  for corruption in _CORRUPTIONS:
+    config_list.append(
+        MNISTCorruptedConfig(
+            name=corruption,
+            version='0.0.1',
+            description='Corruption method: ' + corruption,
+            corruption_type=corruption,
+        ))
+  return config_list
+
+
+class MNISTCorrupted(tfds.core.GeneratorBasedBuilder):
+  """Corrupted MNIST dataset."""
+  BUILDER_CONFIGS = _make_builder_configs()
+
+  def _info(self):
+    """Returns basic information of dataset.
+
+    Returns:
+      tfds.core.DatasetInfo.
+    """
+    return tfds.core.DatasetInfo(
+        builder=self,
+        description=_DESCRIPTION,
+        features=tfds.features.FeaturesDict({
+            'image':
+                tfds.features.Image(shape=mnist.MNIST_IMAGE_SHAPE),
+            'label':
+                tfds.features.ClassLabel(num_classes=mnist.MNIST_NUM_CLASSES),
+        }),
+        supervised_keys=('image', 'label'),
+        urls=['https://github.com/google-research/mnist-c'],
+        citation=_CITATION)
+
+  def _split_generators(self, dl_manager):
+    """Return the train, test split of MNIST-C.
+
+    Args:
+      dl_manager: download manager object.
+
+    Returns:
+      train split, test split.
+    """
+    path = dl_manager.download_and_extract(_DOWNLOAD_URL)
+    return [
+        tfds.core.SplitGenerator(
+            name=tfds.Split.TRAIN,
+            num_shards=1,
+            gen_kwargs={
+                'data_dir': os.path.join(path, _DIRNAME),
+                'is_train': True
+            }),
+        tfds.core.SplitGenerator(
+            name=tfds.Split.TEST,
+            num_shards=1,
+            gen_kwargs={
+                'data_dir': os.path.join(path, _DIRNAME),
+                'is_train': False
+            }),
+    ]
+
+  def _generate_examples(self, data_dir, is_train):
+    """Generate corrupted MNIST data.
+
+    Apply corruptions to the raw images according to self.corruption_type.
+
+    Args:
+      data_dir: root directory of downloaded dataset
+      is_train: whether to return train images or test images
+
+    Yields:
+      dictionary with image file and label.
+    """
+    corruption = self.builder_config.corruption
+
+    if is_train:
+      images_file = os.path.join(data_dir, corruption, _TRAIN_IMAGES_FILENAME)
+      labels_file = os.path.join(data_dir, corruption, _TRAIN_LABELS_FILENAME)
+    else:
+      images_file = os.path.join(data_dir, corruption, _TEST_IMAGES_FILENAME)
+      labels_file = os.path.join(data_dir, corruption, _TEST_LABELS_FILENAME)
+
+    with tf.io.gfile.GFile(labels_file, mode='rb') as f:
+      labels = np.load(f)
+
+    with tf.io.gfile.GFile(images_file, mode='rb') as f:
+      images = np.load(f)
+
+    for image, label in zip(images, labels):
+      yield {
+          'image': image,
+          'label': label,
+      }

tensorflow_datasets/image/mnist_corrupted_test.py

@@ -0,0 +1,37 @@
+# coding=utf-8
+# Copyright 2019 The TensorFlow Datasets Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for corrupted MNIST."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from tensorflow_datasets import testing
+from tensorflow_datasets.image import mnist_corrupted
+
+
+class MNISTCorruptedTest(testing.DatasetBuilderTestCase):
+
+  BUILDER_CONFIG_NAMES_TO_TEST = ["dotted_line"]
+
+  DATASET_CLASS = mnist_corrupted.MNISTCorrupted
+  SPLITS = {
+      "train": 2,
+      "test": 2,
+  }
+
+if __name__ == "__main__":
+  testing.test_main()

tensorflow_datasets/testing/test_data/fake_examples/mnist_corrupted/mnist_c/dotted_line/test_images.npy

@@ -0,0 +1 @@
+https://zenodo.org/record/3239543/files/mnist_c.zip 246661575 af9ee8c6a815870c7fdde5af84c7bf8db0bcfa1f41056db83871037fba70e493