Add the 2017 INaturalist Competition dataset.

PiperOrigin-RevId: 282831191

Author: jpuigcerver

setup.py

@@ -96,6 +96,8 @@
     'image/food-101_classes.txt',
     'image/imagenet2012_labels.txt',
     'image/imagenet2012_validation_labels.txt',
+    'image/inaturalist_labels.txt',
+    'image/inaturalist_supercategories.txt',
     'image/open_images_classes_all.txt',
     'image/open_images_classes_boxable.txt',
     'image/open_images_classes_trainable.txt',

tensorflow_datasets/image/__init__.py

@@ -55,6 +55,7 @@
 from tensorflow_datasets.image.imagenet import Imagenet2012
 from tensorflow_datasets.image.imagenet2012_corrupted import Imagenet2012Corrupted
 from tensorflow_datasets.image.imagenet_resized import ImagenetResized
+from tensorflow_datasets.image.inaturalist import INaturalist2017
 from tensorflow_datasets.image.lfw import LFW
 from tensorflow_datasets.image.lost_and_found import LostAndFound
 from tensorflow_datasets.image.lsun import Lsun

tensorflow_datasets/image/inaturalist.py

@@ -0,0 +1,160 @@
+# 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.
+
+# Lint as: python3
+"""INaturalist datasets."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import json
+import os
+
+import six.moves.urllib as urllib
+import tensorflow as tf
+import tensorflow_datasets.public_api as tfds
+
+_DESCRIPTION = """\
+This dataset contains a total of 5,089 categories, across 579,184 training
+images and 95,986 validation images. For the training set, the distribution of
+images per category follows the observation frequency of that category by the
+iNaturalist community.
+
+Although the original dataset contains some images with bounding boxes,
+currently, only image-level annotations are provided (single label/image).
+In addition, the organizers have not published the test labels, so we only
+provide the test images (label = -1).
+"""
+_CITATION = """\
+@InProceedings{Horn_2018_CVPR,
+author = {
+Van Horn, Grant and Mac Aodha, Oisin and Song, Yang and Cui, Yin and Sun, Chen
+and Shepard, Alex and Adam, Hartwig and Perona, Pietro and Belongie, Serge},
+title = {The INaturalist Species Classification and Detection Dataset},
+booktitle = {
+The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
+month = {June},
+year = {2018}
+}
+"""
+_URL = "http://www.vision.caltech.edu/~gvanhorn/datasets/inaturalist/fgvc4_competition/"
+
+
+class INaturalist2017(tfds.core.GeneratorBasedBuilder):
+  """Dataset from the INaturalist Competition 2017."""
+
+  VERSION = tfds.core.Version("0.1.0")
+
+  def _info(self):
+    """Define the dataset info."""
+    return tfds.core.DatasetInfo(
+        builder=self,
+        description=_DESCRIPTION,
+        features=tfds.features.FeaturesDict({
+            "id": tfds.features.Text(),
+            "image": tfds.features.Image(),
+            "label": tfds.features.ClassLabel(
+                names_file=tfds.core.get_tfds_path(
+                    os.path.join("image", "inaturalist_labels.txt"))),
+            "supercategory": tfds.features.ClassLabel(
+                names_file=tfds.core.get_tfds_path(
+                    os.path.join("image", "inaturalist_supercategories.txt"))),
+        }),
+        supervised_keys=("image", "label"),
+        homepage="https://github.com/visipedia/inat_comp/tree/master/2017",
+        citation=_CITATION)
+
+  def _split_generators(self, dl_manager):
+    output_files = dl_manager.download_and_extract({
+        "trainval_images":
+            tfds.download.Resource(
+                url=urllib.parse.urljoin(_URL, "train_val_images.tar.gz"),
+                extract_method=tfds.download.ExtractMethod.NO_EXTRACT),
+        "trainval_annos":
+            urllib.parse.urljoin(_URL, "train_val2017.zip"),
+        "test_images":
+            tfds.download.Resource(
+                url=urllib.parse.urljoin(_URL, "test2017.tar.gz"),
+                extract_method=tfds.download.ExtractMethod.NO_EXTRACT),
+
+    })
+    return [
+        tfds.core.SplitGenerator(
+            name=tfds.Split.TRAIN,
+            gen_kwargs=dict(
+                images_archive=dl_manager.iter_archive(
+                    output_files["trainval_images"]),
+                annon_file=os.path.join(output_files["trainval_annos"],
+                                        "train2017.json"),
+            ),
+        ),
+        tfds.core.SplitGenerator(
+            name=tfds.Split.VALIDATION,
+            gen_kwargs=dict(
+                images_archive=dl_manager.iter_archive(
+                    output_files["trainval_images"]),
+                annon_file=os.path.join(output_files["trainval_annos"],
+                                        "val2017.json"),
+            ),
+        ),
+        tfds.core.SplitGenerator(
+            name=tfds.Split.TEST,
+            gen_kwargs=dict(
+                images_archive=dl_manager.iter_archive(
+                    output_files["test_images"]),
+                annon_file=None,
+            ),
+        ),
+    ]
+
+  def _generate_examples(self, images_archive, annon_file):
+    """Generate examples."""
+    if annon_file is not None:
+      # Training and validation images.
+      with tf.io.gfile.GFile(annon_file, "r") as f:
+        data = json.load(f)
+      # First read the annotations file, used to filter the contents of the
+      # tar.gz file when yielding examples.
+      key2data = {}
+      for image, annotation in zip(data["images"], data["annotations"]):
+        category_id = annotation["category_id"]
+        category = data["categories"][category_id]["name"]
+        supercategory = data["categories"][category_id]["supercategory"]
+        key = os.path.basename(image["file_name"]).split(".")[0]
+        key2data[key] = {
+            "id": key,
+            "label": category,
+            "supercategory": supercategory,
+        }
+      # Read tar.gz file containing train & validation images and yield relevant
+      # examples.
+      for fpath, fobj in images_archive:
+        key = os.path.basename(fpath).split(".")[0]
+        if key in key2data:
+          data = key2data[key].copy()
+          data["image"] = fobj
+          yield key, data
+    else:
+      # Read tar.gz file containing all test images and yield all examples.
+      for fpath, fobj in images_archive:
+        key = os.path.basename(fpath).split(".")[0]
+        # Note: test labels are not annotated, so just return -1 as labels.
+        yield key, {
+            "id": key,
+            "image": fobj,
+            "label": -1,
+            "supercategory": -1,
+        }