Add so2sat benchmarking dataset with Sentinel-1 and Sentinel-2 measurments and 17 local climate zone (LCZ42) classes.

PiperOrigin-RevId: 249688173

Author: TensorFlow Datasets Team

tensorflow_datasets/image/__init__.py

@@ -51,6 +51,7 @@
 from tensorflow_datasets.image.rock_paper_scissors import RockPaperScissors
 from tensorflow_datasets.image.shapes3d import Shapes3d
 from tensorflow_datasets.image.smallnorb import Smallnorb
+from tensorflow_datasets.image.so2sat import So2sat
 from tensorflow_datasets.image.sun import Sun397
 from tensorflow_datasets.image.svhn import SvhnCropped
 from tensorflow_datasets.image.uc_merced import UcMerced

tensorflow_datasets/image/so2sat.py

@@ -0,0 +1,166 @@
+# 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.
+
+"""So2SAT remote sensing dataset."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import h5py
+import numpy as np
+import tensorflow as tf
+import tensorflow_datasets.public_api as tfds
+
+_DESCRIPTION = """\
+So2Sat LCZ42 is a dataset consisting of co-registered synthetic aperture radar
+and multispectral optical image patches acquired by the Sentinel-1 and
+Sentinel-2 remote sensing satellites, and the corresponding local climate zones
+(LCZ) label. The dataset is distributed over 42 cities across different
+continents and cultural regions of the world.
+
+The full dataset (`all`) consists of 8 Sentinel-1 and 10 Sentinel-2 channels.
+Alternatively, one can select the `rgb` subset, which contains only the optical
+frequency bands of Sentinel-2, rescaled and encoded as JPEG.
+
+Dataset URL: http://doi.org/10.14459/2018MP1454690
+License: http://creativecommons.org/licenses/by/4.0
+"""
+
+_LABELS = [
+    'Compact high-rise', 'Compact mid-rise', 'Compact low-rise',
+    'Open high-rise', 'Open mid-rise', 'Open low-rise', 'Lightweight low-rise',
+    'Large low-rise', 'Sparsely built', 'Heavy industry', 'Dense trees',
+    'Scattered trees', 'Bush or scrub', 'Low plants', 'Bare rock or paved',
+    'Bare soil or sand', 'Water'
+]
+
+_DATA_OPTIONS = ['rgb', 'all']
+
+# Maximal value observed for the optical channels of Sentinel-2 in this dataset.
+_OPTICAL_MAX_VALUE = 2.8
+
+
+class So2satConfig(tfds.core.BuilderConfig):
+  """BuilderConfig for so2sat."""
+
+  def __init__(self, selection=None, **kwargs):
+    """Constructs a So2satConfig.
+
+    Args:
+      selection: `str`, one of `_DATA_OPTIONS`.
+      **kwargs: keyword arguments forwarded to super.
+    """
+    if selection not in _DATA_OPTIONS:
+      raise ValueError('selection must be one of %s' % _DATA_OPTIONS)
+
+    super(So2satConfig, self).__init__(**kwargs)
+    self.selection = selection
+
+
+class So2sat(tfds.core.GeneratorBasedBuilder):
+  """So2SAT remote sensing dataset."""
+
+  BUILDER_CONFIGS = [
+      So2satConfig(
+          selection='rgb',
+          name='rgb',
+          version='0.0.1',
+          description='Sentinel-2 RGB channels'),
+      So2satConfig(
+          selection='all',
+          name='all',
+          version='0.0.1',
+          description='8 Sentinel-1 and 10 Sentinel-2 channels'),
+  ]
+
+  def _info(self):
+    if self.builder_config.selection == 'rgb':
+      features = tfds.features.FeaturesDict({
+          'image': tfds.features.Image(shape=[32, 32, 3]),
+          'label': tfds.features.ClassLabel(names=_LABELS),
+          'sample_id': tfds.features.Tensor(shape=(), dtype=tf.int64),
+      })
+      supervised_keys = ('image', 'label')
+    elif self.builder_config.selection == 'all':
+      features = tfds.features.FeaturesDict({
+          'sentinel1':
+              tfds.features.Tensor(shape=[32, 32, 8], dtype=tf.float32),
+          'sentinel2':
+              tfds.features.Tensor(shape=[32, 32, 10], dtype=tf.float32),
+          'label':
+              tfds.features.ClassLabel(names=_LABELS),
+          'sample_id':
+              tfds.features.Tensor(shape=(), dtype=tf.int64),
+      })
+      supervised_keys = None
+    return tfds.core.DatasetInfo(
+        builder=self,
+        description=_DESCRIPTION,
+        features=features,
+        supervised_keys=supervised_keys,
+        urls=['http://doi.org/10.14459/2018MP1454690'],
+    )
+
+  def _split_generators(self, dl_manager):
+    """Returns SplitGenerators."""
+    paths = dl_manager.download({
+        'train': 'ftp://m1454690:m1454690@dataserv.ub.tum.de/training.h5',
+        'val': 'ftp://m1454690:m1454690@dataserv.ub.tum.de/validation.h5'
+    })
+    return [
+        tfds.core.SplitGenerator(
+            name=tfds.Split.TRAIN,
+            num_shards=20,
+            gen_kwargs={
+                'path': paths['train'],
+                'selection': self.builder_config.selection,
+            },
+        ),
+        tfds.core.SplitGenerator(
+            name=tfds.Split.VALIDATION,
+            num_shards=5,
+            gen_kwargs={
+                'path': paths['val'],
+                'selection': self.builder_config.selection,
+            },
+        ),
+    ]
+
+  def _generate_examples(self, path, selection):
+    """Yields examples."""
+    with h5py.File(path, 'r') as fid:
+      sen1 = fid['sen1']
+      sen2 = fid['sen2']
+      label = fid['label']
+      for i in range(len(sen1)):
+        if selection == 'rgb':
+          yield {
+              'image': _create_rgb(sen2[i]),
+              'label': np.argmax(label[i]).astype(int),
+              'sample_id': i,
+          }
+        elif selection == 'all':
+          yield {
+              'sentinel1': sen1[i].astype(np.float32),
+              'sentinel2': sen2[i].astype(np.float32),
+              'label': np.argmax(label[i]).astype(int),
+              'sample_id': i,
+          }
+
+
+def _create_rgb(sen2_bands):
+  return np.clip(sen2_bands[..., [2, 1, 0]] / _OPTICAL_MAX_VALUE * 255.0, 0,
+                 255).astype(np.uint8)

tensorflow_datasets/image/so2sat_test.py

@@ -0,0 +1,39 @@
+# 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.
+
+"""TODO(so2sat): Add a description here."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from tensorflow_datasets import testing
+from tensorflow_datasets.image import so2sat
+
+
+class So2satTest(testing.DatasetBuilderTestCase):
+  DATASET_CLASS = so2sat.So2sat
+  SPLITS = {
+      "train": 5,  # Number of fake train example
+      "validation": 3,  # Number of fake validation example
+  }
+  DL_EXTRACT_RESULT = {
+      "train": "./training.h5",
+      "val": "./validation.h5",
+  }
+
+
+if __name__ == "__main__":
+  testing.test_main()

tensorflow_datasets/testing/test_data/fake_examples/so2sat/training.h5

@@ -0,0 +1,2 @@
+ftp://m1454690:m1454690@dataserv.ub.tum.de/training.h5 52006404720 06df6c9b8875e37f172ba548f466640293d77ad3be335c2d5dfcba3d35942daf
+ftp://m1454690:m1454690@dataserv.ub.tum.de/validation.h5 3559773496 cab820b5176a6b5fb35ab423f434e40b073265a7b6317d9f6895a9fa7c0bb285