Release MobileDet code and model, and require tf_slim installation for OD API. (#8562)

* Merged commit includes the following changes:
311933687  by Sergio Guadarrama:

    Removes spurios use of tf.compat.v2, which results in spurious tf.compat.v1.compat.v2. Adds basic test to nasnet_utils.
    Replaces all remaining import tensorflow as tf with import tensorflow.compat.v1 as tf

--
311766063  by Sergio Guadarrama:

    Removes explicit tf.compat.v1 in all call sites (we already import tf.compat.v1, so this code was  doing tf.compat.v1.compat.v1). The existing code worked in latest version of tensorflow, 2.2, (and 1.15) but not in 1.14 or in 2.0.0a, this CL fixes it.

--
311624958  by Sergio Guadarrama:

    Updates README that doesn't render properly in github documentation

--
310980959  by Sergio Guadarrama:

    Moves research_models/slim off tf.contrib.slim/layers/framework to tf_slim

--
310263156  by Sergio Guadarrama:

    Adds model breakdown for MobilenetV3

--
308640516  by Sergio Guadarrama:

    Internal change

308244396  by Sergio Guadarrama:

    GroupNormalization support for MobilenetV3.

--
307475800  by Sergio Guadarrama:

    Internal change

--
302077708  by Sergio Guadarrama:

    Remove `disable_tf2` behavior from slim py_library targets

--
301208453  by Sergio Guadarrama:

    Automated refactoring to make code Python 3 compatible.

--
300816672  by Sergio Guadarrama:

    Internal change

299433840  by Sergio Guadarrama:

    Internal change

299221609  by Sergio Guadarrama:

    Explicitly disable Tensorflow v2 behaviors for all TF1.x binaries and tests

--
299179617  by Sergio Guadarrama:

    Internal change

299040784  by Sergio Guadarrama:

    Internal change

299036699  by Sergio Guadarrama:

    Internal change

298736510  by Sergio Guadarrama:

    Internal change

298732599  by Sergio Guadarrama:

    Internal change

298729507  by Sergio Guadarrama:

    Internal change

298253328  by Sergio Guadarrama:

    Internal change

297788346  by Sergio Guadarrama:

    Internal change

297785278  by Sergio Guadarrama:

    Internal change

297783127  by Sergio Guadarrama:

    Internal change

297725870  by Sergio Guadarrama:

    Internal change

297721811  by Sergio Guadarrama:

    Internal change

297711347  by Sergio Guadarrama:

    Internal change

297708059  by Sergio Guadarrama:

    Internal change

297701831  by Sergio Guadarrama:

    Internal change

297700038  by Sergio Guadarrama:

    Internal change

297670468  by Sergio Guadarrama:

    Internal change.

--
297350326  by Sergio Guadarrama:

    Explicitly replace "import tensorflow" with "tensorflow.compat.v1" for TF2.x migration

--
297201668  by Sergio Guadarrama:

    Explicitly replace "import tensorflow" with "tensorflow.compat.v1" for TF2.x migration

--
294483372  by Sergio Guadarrama:

    Internal change

PiperOrigin-RevId: 311933687

* Merged commit includes the following changes:
312578615  by Menglong Zhu:

    Modify the LSTM feature extractors to be python 3 compatible.

--
311264357  by Menglong Zhu:

    Removes contrib.slim

--
308957207  by Menglong Zhu:

    Automated refactoring to make code Python 3 compatible.

--
306976470  by yongzhe:

    Internal change

306777559  by Menglong Zhu:

    Internal change

--
299232507  by lzyuan:

    Internal update.

--
299221735  by lzyuan:

    Add small epsilon on max_range for quantize_op to prevent range collapse.

--

PiperOrigin-RevId: 312578615

* Merged commit includes the following changes:
310447280  by lzc:

    Internal changes.

--

PiperOrigin-RevId: 310447280
Co-authored-by: Sergio Guadarrama <sguada@google.com>
Co-authored-by: Menglong Zhu <menglong@google.com>

research/object_detection/data_decoders/tf_sequence_example_decoder_test.py

+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# 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 tf_sequence_example_decoder.py."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import numpy as np
+import tensorflow.compat.v1 as tf
+
+from object_detection.core import standard_fields as fields
+from object_detection.data_decoders import tf_sequence_example_decoder
+from object_detection.dataset_tools import seq_example_util
+from object_detection.utils import test_case
+
+
+class TfSequenceExampleDecoderTest(test_case.TestCase):
+
+  def _create_label_map(self, path):
+    label_map_text = """
+      item {
+        name: "dog"
+        id: 1
+      }
+      item {
+        name: "cat"
+        id: 2
+      }
+      item {
+        name: "panda"
+        id: 4
+      }
+    """
+    with tf.gfile.Open(path, 'wb') as f:
+      f.write(label_map_text)
+
+  def _make_random_serialized_jpeg_images(self, num_frames, image_height,
+                                          image_width):
+    def graph_fn():
+      images = tf.cast(tf.random.uniform(
+          [num_frames, image_height, image_width, 3],
+          maxval=256,
+          dtype=tf.int32), dtype=tf.uint8)
+      images_list = tf.unstack(images, axis=0)
+      return [tf.io.encode_jpeg(image) for image in images_list]
+    encoded_images = self.execute(graph_fn, [])
+    return encoded_images
+
+  def test_decode_sequence_example(self):
+    num_frames = 4
+    image_height = 20
+    image_width = 30
+
+    expected_groundtruth_boxes = [
+        [[0.0, 0.0, 1.0, 1.0], [0.0, 0.0, 0.0, 0.0]],
+        [[0.2, 0.2, 1.0, 1.0], [0.0, 0.0, 1.0, 1.0]],
+        [[0.0, 0.0, 1.0, 1.0], [0.1, 0.1, 0.2, 0.2]],
+        [[0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0]]
+    ]
+    expected_groundtruth_classes = [
+        [-1, -1],
+        [-1, 1],
+        [1, 2],
+        [-1, -1]
+    ]
+
+    flds = fields.InputDataFields
+    encoded_images = self._make_random_serialized_jpeg_images(
+        num_frames, image_height, image_width)
+
+    def graph_fn():
+      label_map_proto_file = os.path.join(self.get_temp_dir(), 'labelmap.pbtxt')
+      self._create_label_map(label_map_proto_file)
+      decoder = tf_sequence_example_decoder.TfSequenceExampleDecoder(
+          label_map_proto_file=label_map_proto_file)
+      sequence_example_serialized = seq_example_util.make_sequence_example(
+          dataset_name='video_dataset',
+          video_id='video',
+          encoded_images=encoded_images,
+          image_height=image_height,
+          image_width=image_width,
+          image_format='JPEG',
+          image_source_ids=[str(i) for i in range(num_frames)],
+          is_annotated=[[1], [1], [1], [1]],
+          bboxes=[
+              [[0., 0., 1., 1.]],  # Frame 0.
+              [[0.2, 0.2, 1., 1.],
+               [0., 0., 1., 1.]],  # Frame 1.
+              [[0., 0., 1., 1.],  # Frame 2.
+               [0.1, 0.1, 0.2, 0.2]],
+              [[]],  # Frame 3.
+          ],
+          label_strings=[
+              ['fox'],  # Frame 0. Fox will be filtered out.
+              ['fox', 'dog'],  # Frame 1. Fox will be filtered out.
+              ['dog', 'cat'],  # Frame 2.
+              [],  # Frame 3
+          ]).SerializeToString()
+
+      example_string_tensor = tf.convert_to_tensor(sequence_example_serialized)
+      return decoder.decode(example_string_tensor)
+
+    tensor_dict_out = self.execute(graph_fn, [])
+    self.assertAllClose(expected_groundtruth_boxes,
+                        tensor_dict_out[flds.groundtruth_boxes])
+    self.assertAllEqual(expected_groundtruth_classes,
+                        tensor_dict_out[flds.groundtruth_classes])
+
+  def test_decode_sequence_example_negative_clip(self):
+    num_frames = 4
+    image_height = 20
+    image_width = 30
+
+    expected_groundtruth_boxes = -1 * np.ones((4, 0, 4))
+    expected_groundtruth_classes = -1 * np.ones((4, 0))
+
+    flds = fields.InputDataFields
+
+    encoded_images = self._make_random_serialized_jpeg_images(
+        num_frames, image_height, image_width)
+
+    def graph_fn():
+      sequence_example_serialized = seq_example_util.make_sequence_example(
+          dataset_name='video_dataset',
+          video_id='video',
+          encoded_images=encoded_images,
+          image_height=image_height,
+          image_width=image_width,
+          image_format='JPEG',
+          image_source_ids=[str(i) for i in range(num_frames)],
+          bboxes=[
+              [[]],
+              [[]],
+              [[]],
+              [[]]
+          ],
+          label_strings=[
+              [],
+              [],
+              [],
+              []
+          ]).SerializeToString()
+      example_string_tensor = tf.convert_to_tensor(sequence_example_serialized)
+
+      label_map_proto_file = os.path.join(self.get_temp_dir(), 'labelmap.pbtxt')
+      self._create_label_map(label_map_proto_file)
+      decoder = tf_sequence_example_decoder.TfSequenceExampleDecoder(
+          label_map_proto_file=label_map_proto_file)
+      return decoder.decode(example_string_tensor)
+
+    tensor_dict_out = self.execute(graph_fn, [])
+    self.assertAllClose(expected_groundtruth_boxes,
+                        tensor_dict_out[flds.groundtruth_boxes])
+    self.assertAllEqual(expected_groundtruth_classes,
+                        tensor_dict_out[flds.groundtruth_classes])
+
+
+if __name__ == '__main__':
+  tf.test.main()

research/object_detection/dataset_tools/create_coco_tf_record.py

@@ -40,7 +40,7 @@ import numpy as np
 import PIL.Image
 
 from pycocotools import mask
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.dataset_tools import tf_record_creation_util
 from object_detection.utils import dataset_util

research/object_detection/dataset_tools/create_coco_tf_record_test.py

@@ -21,7 +21,7 @@ import os
 import numpy as np
 import PIL.Image
 import six
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.dataset_tools import create_coco_tf_record
 

research/object_detection/dataset_tools/create_kitti_tf_record.py

@@ -42,7 +42,7 @@ import os
 
 import numpy as np
 import PIL.Image as pil
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.utils import dataset_util
 from object_detection.utils import label_map_util

research/object_detection/dataset_tools/create_kitti_tf_record_test.py

@@ -20,7 +20,7 @@ import os
 import numpy as np
 import PIL.Image
 import six
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.dataset_tools import create_kitti_tf_record
 

research/object_detection/dataset_tools/create_oid_tf_record.py

@@ -40,7 +40,7 @@ import os
 
 import contextlib2
 import pandas as pd
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.dataset_tools import oid_tfrecord_creation
 from object_detection.dataset_tools import tf_record_creation_util

research/object_detection/dataset_tools/create_pascal_tf_record.py

@@ -32,7 +32,7 @@ import os
 
 from lxml import etree
 import PIL.Image
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.utils import dataset_util
 from object_detection.utils import label_map_util

research/object_detection/dataset_tools/create_pascal_tf_record_test.py

@@ -20,7 +20,7 @@ import os
 import numpy as np
 import PIL.Image
 import six
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.dataset_tools import create_pascal_tf_record
 

research/object_detection/dataset_tools/create_pet_tf_record.py

@@ -37,7 +37,7 @@ import contextlib2
 from lxml import etree
 import numpy as np
 import PIL.Image
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.dataset_tools import tf_record_creation_util
 from object_detection.utils import dataset_util

research/object_detection/dataset_tools/oid_hierarchical_labels_expansion_test.py

@@ -18,7 +18,7 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.dataset_tools import oid_hierarchical_labels_expansion
 

research/object_detection/dataset_tools/oid_tfrecord_creation.py

@@ -19,7 +19,7 @@ from __future__ import division
 from __future__ import print_function
 
 import six
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.core import standard_fields
 from object_detection.utils import dataset_util

research/object_detection/dataset_tools/oid_tfrecord_creation_test.py

@@ -16,7 +16,7 @@
 
 import pandas as pd
 import six
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.dataset_tools import oid_tfrecord_creation
 

research/object_detection/dataset_tools/seq_example_util.py

+# Lint as: python2, python3
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# 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.
+# ==============================================================================
+"""Common utility for object detection tf.train.SequenceExamples."""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import tensorflow.compat.v1 as tf
+
+
+def context_float_feature(ndarray):
+  """Converts a numpy float array to a context float feature.
+
+  Args:
+    ndarray: A numpy float array.
+
+  Returns:
+    A context float feature.
+  """
+  feature = tf.train.Feature()
+  for val in ndarray:
+    feature.float_list.value.append(val)
+  return feature
+
+
+def context_int64_feature(ndarray):
+  """Converts a numpy array to a context int64 feature.
+
+  Args:
+    ndarray: A numpy int64 array.
+
+  Returns:
+    A context int64 feature.
+  """
+  feature = tf.train.Feature()
+  for val in ndarray:
+    feature.int64_list.value.append(val)
+  return feature
+
+
+def context_bytes_feature(ndarray):
+  """Converts a numpy bytes array to a context bytes feature.
+
+  Args:
+    ndarray: A numpy bytes array.
+
+  Returns:
+    A context bytes feature.
+  """
+  feature = tf.train.Feature()
+  for val in ndarray:
+    if isinstance(val, np.ndarray):
+      val = val.tolist()
+    feature.bytes_list.value.append(tf.compat.as_bytes(val))
+  return feature
+
+
+def sequence_float_feature(ndarray):
+  """Converts a numpy float array to a sequence float feature.
+
+  Args:
+    ndarray: A numpy float array.
+
+  Returns:
+    A sequence float feature.
+  """
+  feature_list = tf.train.FeatureList()
+  for row in ndarray:
+    feature = feature_list.feature.add()
+    if row.size:
+      feature.float_list.value[:] = row
+  return feature_list
+
+
+def sequence_int64_feature(ndarray):
+  """Converts a numpy int64 array to a sequence int64 feature.
+
+  Args:
+    ndarray: A numpy int64 array.
+
+  Returns:
+    A sequence int64 feature.
+  """
+  feature_list = tf.train.FeatureList()
+  for row in ndarray:
+    feature = feature_list.feature.add()
+    if row.size:
+      feature.int64_list.value[:] = row
+  return feature_list
+
+
+def sequence_bytes_feature(ndarray):
+  """Converts a bytes float array to a sequence bytes feature.
+
+  Args:
+    ndarray: A numpy bytes array.
+
+  Returns:
+    A sequence bytes feature.
+  """
+  feature_list = tf.train.FeatureList()
+  for row in ndarray:
+    if isinstance(row, np.ndarray):
+      row = row.tolist()
+    feature = feature_list.feature.add()
+    if row:
+      row = [tf.compat.as_bytes(val) for val in row]
+      feature.bytes_list.value[:] = row
+  return feature_list
+
+
+def boxes_to_box_components(bboxes):
+  """Converts a list of numpy arrays (boxes) to box components.
+
+  Args:
+    bboxes: A numpy array of bounding boxes.
+
+  Returns:
+    Bounding box component lists.
+  """
+  ymin_list = []
+  xmin_list = []
+  ymax_list = []
+  xmax_list = []
+  for bbox in bboxes:
+    bbox = np.array(bbox).astype(np.float32)
+    ymin, xmin, ymax, xmax = np.split(bbox, 4, axis=1)
+    ymin_list.append(np.reshape(ymin, [-1]))
+    xmin_list.append(np.reshape(xmin, [-1]))
+    ymax_list.append(np.reshape(ymax, [-1]))
+    xmax_list.append(np.reshape(xmax, [-1]))
+  return ymin_list, xmin_list, ymax_list, xmax_list
+
+
+def make_sequence_example(dataset_name,
+                          video_id,
+                          encoded_images,
+                          image_height,
+                          image_width,
+                          image_format=None,
+                          image_source_ids=None,
+                          timestamps=None,
+                          is_annotated=None,
+                          bboxes=None,
+                          label_strings=None,
+                          detection_bboxes=None,
+                          detection_classes=None,
+                          detection_scores=None):
+  """Constructs tf.SequenceExamples.
+
+  Args:
+    dataset_name: String with dataset name.
+    video_id: String with video id.
+    encoded_images: A [num_frames] list (or numpy array) of encoded image
+      frames.
+    image_height: Height of the images.
+    image_width: Width of the images.
+    image_format: Format of encoded images.
+    image_source_ids: (Optional) A [num_frames] list of unique string ids for
+      each image.
+    timestamps: (Optional) A [num_frames] list (or numpy array) array with image
+      timestamps.
+    is_annotated: (Optional) A [num_frames] list (or numpy array) array
+      in which each element indicates whether the frame has been annotated
+      (1) or not (0).
+    bboxes: (Optional) A list (with num_frames elements) of [num_boxes_i, 4]
+      numpy float32 arrays holding boxes for each frame.
+    label_strings: (Optional) A list (with num_frames_elements) of [num_boxes_i]
+      numpy string arrays holding object string labels for each frame.
+    detection_bboxes: (Optional) A list (with num_frames elements) of
+      [num_boxes_i, 4] numpy float32 arrays holding prediction boxes for each
+      frame.
+    detection_classes: (Optional) A list (with num_frames_elements) of
+      [num_boxes_i] numpy int64 arrays holding predicted classes for each frame.
+    detection_scores: (Optional) A list (with num_frames_elements) of
+      [num_boxes_i] numpy float32 arrays holding predicted object scores for
+      each frame.
+
+  Returns:
+    A tf.train.SequenceExample.
+  """
+  num_frames = len(encoded_images)
+  image_encoded = np.expand_dims(encoded_images, axis=-1)
+  if timestamps is None:
+    timestamps = np.arange(num_frames)
+  image_timestamps = np.expand_dims(timestamps, axis=-1)
+
+  # Context fields.
+  context_dict = {
+      'example/dataset_name': context_bytes_feature([dataset_name]),
+      'clip/start/timestamp': context_int64_feature([image_timestamps[0][0]]),
+      'clip/end/timestamp': context_int64_feature([image_timestamps[-1][0]]),
+      'clip/frames': context_int64_feature([num_frames]),
+      'image/channels': context_int64_feature([3]),
+      'image/height': context_int64_feature([image_height]),
+      'image/width': context_int64_feature([image_width]),
+      'clip/media_id': context_bytes_feature([video_id])
+  }
+
+  # Sequence fields.
+  feature_list = {
+      'image/encoded': sequence_bytes_feature(image_encoded),
+      'image/timestamp': sequence_int64_feature(image_timestamps),
+  }
+
+  # Add optional fields.
+  if image_format is not None:
+    context_dict['image/format'] = context_bytes_feature([image_format])
+  if image_source_ids is not None:
+    feature_list['image/source_id'] = sequence_bytes_feature(image_source_ids)
+  if bboxes is not None:
+    bbox_ymin, bbox_xmin, bbox_ymax, bbox_xmax = boxes_to_box_components(bboxes)
+    feature_list['region/bbox/xmin'] = sequence_float_feature(bbox_xmin)
+    feature_list['region/bbox/xmax'] = sequence_float_feature(bbox_xmax)
+    feature_list['region/bbox/ymin'] = sequence_float_feature(bbox_ymin)
+    feature_list['region/bbox/ymax'] = sequence_float_feature(bbox_ymax)
+    if is_annotated is None:
+      is_annotated = np.ones(num_frames, dtype=np.int64)
+    is_annotated = np.expand_dims(is_annotated, axis=-1)
+    feature_list['region/is_annotated'] = sequence_int64_feature(is_annotated)
+
+  if label_strings is not None:
+    feature_list['region/label/string'] = sequence_bytes_feature(
+        label_strings)
+
+  if detection_bboxes is not None:
+    det_bbox_ymin, det_bbox_xmin, det_bbox_ymax, det_bbox_xmax = (
+        boxes_to_box_components(detection_bboxes))
+    feature_list['predicted/region/bbox/xmin'] = sequence_float_feature(
+        det_bbox_xmin)
+    feature_list['predicted/region/bbox/xmax'] = sequence_float_feature(
+        det_bbox_xmax)
+    feature_list['predicted/region/bbox/ymin'] = sequence_float_feature(
+        det_bbox_ymin)
+    feature_list['predicted/region/bbox/ymax'] = sequence_float_feature(
+        det_bbox_ymax)
+  if detection_classes is not None:
+    feature_list['predicted/region/label/index'] = sequence_int64_feature(
+        detection_classes)
+  if detection_scores is not None:
+    feature_list['predicted/region/label/confidence'] = sequence_float_feature(
+        detection_scores)
+
+  context = tf.train.Features(feature=context_dict)
+  feature_lists = tf.train.FeatureLists(feature_list=feature_list)
+
+  sequence_example = tf.train.SequenceExample(
+      context=context,
+      feature_lists=feature_lists)
+  return sequence_example

research/object_detection/dataset_tools/seq_example_util_test.py

+# Lint as: python2, python3
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# 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 object_detection.utils.seq_example_util."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import six
+import tensorflow.compat.v1 as tf
+
+from object_detection.dataset_tools import seq_example_util
+
+
+class SeqExampleUtilTest(tf.test.TestCase):
+
+  def test_make_unlabeled_example(self):
+    num_frames = 5
+    image_height = 100
+    image_width = 200
+    dataset_name = b'unlabeled_dataset'
+    video_id = b'video_000'
+    images = tf.cast(tf.random.uniform(
+        [num_frames, image_height, image_width, 3],
+        maxval=256,
+        dtype=tf.int32), dtype=tf.uint8)
+    image_source_ids = [str(idx) for idx in range(num_frames)]
+    images_list = tf.unstack(images, axis=0)
+    encoded_images_list = [tf.io.encode_jpeg(image) for image in images_list]
+    with tf.Session() as sess:
+      encoded_images = sess.run(encoded_images_list)
+    seq_example = seq_example_util.make_sequence_example(
+        dataset_name=dataset_name,
+        video_id=video_id,
+        encoded_images=encoded_images,
+        image_height=image_height,
+        image_width=image_width,
+        image_format='JPEG',
+        image_source_ids=image_source_ids)
+
+    context_feature_dict = seq_example.context.feature
+    self.assertEqual(
+        dataset_name,
+        context_feature_dict['example/dataset_name'].bytes_list.value[0])
+    self.assertEqual(
+        0,
+        context_feature_dict['clip/start/timestamp'].int64_list.value[0])
+    self.assertEqual(
+        num_frames - 1,
+        context_feature_dict['clip/end/timestamp'].int64_list.value[0])
+    self.assertEqual(
+        num_frames,
+        context_feature_dict['clip/frames'].int64_list.value[0])
+    self.assertEqual(
+        3,
+        context_feature_dict['image/channels'].int64_list.value[0])
+    self.assertEqual(
+        b'JPEG',
+        context_feature_dict['image/format'].bytes_list.value[0])
+    self.assertEqual(
+        image_height,
+        context_feature_dict['image/height'].int64_list.value[0])
+    self.assertEqual(
+        image_width,
+        context_feature_dict['image/width'].int64_list.value[0])
+    self.assertEqual(
+        video_id,
+        context_feature_dict['clip/media_id'].bytes_list.value[0])
+
+    seq_feature_dict = seq_example.feature_lists.feature_list
+    self.assertLen(
+        seq_feature_dict['image/encoded'].feature[:],
+        num_frames)
+    timestamps = [
+        feature.int64_list.value[0] for feature
+        in seq_feature_dict['image/timestamp'].feature]
+    self.assertAllEqual(list(range(num_frames)), timestamps)
+    source_ids = [
+        feature.bytes_list.value[0] for feature
+        in seq_feature_dict['image/source_id'].feature]
+    self.assertAllEqual(
+        [six.ensure_binary(str(idx)) for idx in range(num_frames)],
+        source_ids)
+
+  def test_make_labeled_example(self):
+    num_frames = 2
+    image_height = 100
+    image_width = 200
+    dataset_name = b'unlabeled_dataset'
+    video_id = b'video_000'
+    labels = [b'dog', b'cat']
+    images = tf.cast(tf.random.uniform(
+        [num_frames, image_height, image_width, 3],
+        maxval=256,
+        dtype=tf.int32), dtype=tf.uint8)
+    images_list = tf.unstack(images, axis=0)
+    encoded_images_list = [tf.io.encode_jpeg(image) for image in images_list]
+    with tf.Session() as sess:
+      encoded_images = sess.run(encoded_images_list)
+    timestamps = [100000, 110000]
+    is_annotated = [1, 0]
+    bboxes = [
+        np.array([[0., 0., 0., 0.],
+                  [0., 0., 1., 1.]], dtype=np.float32),
+        np.zeros([0, 4], dtype=np.float32)
+    ]
+    label_strings = [
+        np.array(labels),
+        np.array([])
+    ]
+
+    seq_example = seq_example_util.make_sequence_example(
+        dataset_name=dataset_name,
+        video_id=video_id,
+        encoded_images=encoded_images,
+        image_height=image_height,
+        image_width=image_width,
+        timestamps=timestamps,
+        is_annotated=is_annotated,
+        bboxes=bboxes,
+        label_strings=label_strings)
+
+    context_feature_dict = seq_example.context.feature
+    self.assertEqual(
+        dataset_name,
+        context_feature_dict['example/dataset_name'].bytes_list.value[0])
+    self.assertEqual(
+        timestamps[0],
+        context_feature_dict['clip/start/timestamp'].int64_list.value[0])
+    self.assertEqual(
+        timestamps[-1],
+        context_feature_dict['clip/end/timestamp'].int64_list.value[0])
+    self.assertEqual(
+        num_frames,
+        context_feature_dict['clip/frames'].int64_list.value[0])
+
+    seq_feature_dict = seq_example.feature_lists.feature_list
+    self.assertLen(
+        seq_feature_dict['image/encoded'].feature[:],
+        num_frames)
+    actual_timestamps = [
+        feature.int64_list.value[0] for feature
+        in seq_feature_dict['image/timestamp'].feature]
+    self.assertAllEqual(timestamps, actual_timestamps)
+    # Frame 0.
+    self.assertAllEqual(
+        is_annotated[0],
+        seq_feature_dict['region/is_annotated'].feature[0].int64_list.value[0])
+    self.assertAllClose(
+        [0., 0.],
+        seq_feature_dict['region/bbox/ymin'].feature[0].float_list.value[:])
+    self.assertAllClose(
+        [0., 0.],
+        seq_feature_dict['region/bbox/xmin'].feature[0].float_list.value[:])
+    self.assertAllClose(
+        [0., 1.],
+        seq_feature_dict['region/bbox/ymax'].feature[0].float_list.value[:])
+    self.assertAllClose(
+        [0., 1.],
+        seq_feature_dict['region/bbox/xmax'].feature[0].float_list.value[:])
+    self.assertAllEqual(
+        labels,
+        seq_feature_dict['region/label/string'].feature[0].bytes_list.value[:])
+
+    # Frame 1.
+    self.assertAllEqual(
+        is_annotated[1],
+        seq_feature_dict['region/is_annotated'].feature[1].int64_list.value[0])
+    self.assertAllClose(
+        [],
+        seq_feature_dict['region/bbox/ymin'].feature[1].float_list.value[:])
+    self.assertAllClose(
+        [],
+        seq_feature_dict['region/bbox/xmin'].feature[1].float_list.value[:])
+    self.assertAllClose(
+        [],
+        seq_feature_dict['region/bbox/ymax'].feature[1].float_list.value[:])
+    self.assertAllClose(
+        [],
+        seq_feature_dict['region/bbox/xmax'].feature[1].float_list.value[:])
+    self.assertAllEqual(
+        [],
+        seq_feature_dict['region/label/string'].feature[1].bytes_list.value[:])
+
+  def test_make_labeled_example_with_predictions(self):
+    num_frames = 2
+    image_height = 100
+    image_width = 200
+    dataset_name = b'unlabeled_dataset'
+    video_id = b'video_000'
+    images = tf.cast(tf.random.uniform(
+        [num_frames, image_height, image_width, 3],
+        maxval=256,
+        dtype=tf.int32), dtype=tf.uint8)
+    images_list = tf.unstack(images, axis=0)
+    encoded_images_list = [tf.io.encode_jpeg(image) for image in images_list]
+    with tf.Session() as sess:
+      encoded_images = sess.run(encoded_images_list)
+    bboxes = [
+        np.array([[0., 0., 0.75, 0.75],
+                  [0., 0., 1., 1.]], dtype=np.float32),
+        np.array([[0., 0.25, 0.5, 0.75]], dtype=np.float32)
+    ]
+    label_strings = [
+        np.array(['cat', 'frog']),
+        np.array(['cat'])
+    ]
+    detection_bboxes = [
+        np.array([[0., 0., 0.75, 0.75]], dtype=np.float32),
+        np.zeros([0, 4], dtype=np.float32)
+    ]
+    detection_classes = [
+        np.array([5], dtype=np.int64),
+        np.array([], dtype=np.int64)
+    ]
+    detection_scores = [
+        np.array([0.9], dtype=np.float32),
+        np.array([], dtype=np.float32)
+    ]
+
+    seq_example = seq_example_util.make_sequence_example(
+        dataset_name=dataset_name,
+        video_id=video_id,
+        encoded_images=encoded_images,
+        image_height=image_height,
+        image_width=image_width,
+        bboxes=bboxes,
+        label_strings=label_strings,
+        detection_bboxes=detection_bboxes,
+        detection_classes=detection_classes,
+        detection_scores=detection_scores)
+
+    context_feature_dict = seq_example.context.feature
+    self.assertEqual(
+        dataset_name,
+        context_feature_dict['example/dataset_name'].bytes_list.value[0])
+    self.assertEqual(
+        0,
+        context_feature_dict['clip/start/timestamp'].int64_list.value[0])
+    self.assertEqual(
+        1,
+        context_feature_dict['clip/end/timestamp'].int64_list.value[0])
+    self.assertEqual(
+        num_frames,
+        context_feature_dict['clip/frames'].int64_list.value[0])
+
+    seq_feature_dict = seq_example.feature_lists.feature_list
+    self.assertLen(
+        seq_feature_dict['image/encoded'].feature[:],
+        num_frames)
+    actual_timestamps = [
+        feature.int64_list.value[0] for feature
+        in seq_feature_dict['image/timestamp'].feature]
+    self.assertAllEqual([0, 1], actual_timestamps)
+    # Frame 0.
+    self.assertAllEqual(
+        1,
+        seq_feature_dict['region/is_annotated'].feature[0].int64_list.value[0])
+    self.assertAllClose(
+        [0., 0.],
+        seq_feature_dict['region/bbox/ymin'].feature[0].float_list.value[:])
+    self.assertAllClose(
+        [0., 0.],
+        seq_feature_dict['region/bbox/xmin'].feature[0].float_list.value[:])
+    self.assertAllClose(
+        [0.75, 1.],
+        seq_feature_dict['region/bbox/ymax'].feature[0].float_list.value[:])
+    self.assertAllClose(
+        [0.75, 1.],
+        seq_feature_dict['region/bbox/xmax'].feature[0].float_list.value[:])
+    self.assertAllEqual(
+        ['cat', 'frog'],
+        seq_feature_dict['region/label/string'].feature[0].bytes_list.value[:])
+    self.assertAllClose(
+        [0.],
+        seq_feature_dict[
+            'predicted/region/bbox/ymin'].feature[0].float_list.value[:])
+    self.assertAllClose(
+        [0.],
+        seq_feature_dict[
+            'predicted/region/bbox/xmin'].feature[0].float_list.value[:])
+    self.assertAllClose(
+        [0.75],
+        seq_feature_dict[
+            'predicted/region/bbox/ymax'].feature[0].float_list.value[:])
+    self.assertAllClose(
+        [0.75],
+        seq_feature_dict[
+            'predicted/region/bbox/xmax'].feature[0].float_list.value[:])
+    self.assertAllEqual(
+        [5],
+        seq_feature_dict[
+            'predicted/region/label/index'].feature[0].int64_list.value[:])
+    self.assertAllClose(
+        [0.9],
+        seq_feature_dict[
+            'predicted/region/label/confidence'].feature[0].float_list.value[:])
+
+    # Frame 1.
+    self.assertAllEqual(
+        1,
+        seq_feature_dict['region/is_annotated'].feature[1].int64_list.value[0])
+    self.assertAllClose(
+        [0.0],
+        seq_feature_dict['region/bbox/ymin'].feature[1].float_list.value[:])
+    self.assertAllClose(
+        [0.25],
+        seq_feature_dict['region/bbox/xmin'].feature[1].float_list.value[:])
+    self.assertAllClose(
+        [0.5],
+        seq_feature_dict['region/bbox/ymax'].feature[1].float_list.value[:])
+    self.assertAllClose(
+        [0.75],
+        seq_feature_dict['region/bbox/xmax'].feature[1].float_list.value[:])
+    self.assertAllEqual(
+        ['cat'],
+        seq_feature_dict['region/label/string'].feature[1].bytes_list.value[:])
+    self.assertAllClose(
+        [],
+        seq_feature_dict[
+            'predicted/region/bbox/ymin'].feature[1].float_list.value[:])
+    self.assertAllClose(
+        [],
+        seq_feature_dict[
+            'predicted/region/bbox/xmin'].feature[1].float_list.value[:])
+    self.assertAllClose(
+        [],
+        seq_feature_dict[
+            'predicted/region/bbox/ymax'].feature[1].float_list.value[:])
+    self.assertAllClose(
+        [],
+        seq_feature_dict[
+            'predicted/region/bbox/xmax'].feature[1].float_list.value[:])
+    self.assertAllEqual(
+        [],
+        seq_feature_dict[
+            'predicted/region/label/index'].feature[1].int64_list.value[:])
+    self.assertAllClose(
+        [],
+        seq_feature_dict[
+            'predicted/region/label/confidence'].feature[1].float_list.value[:])
+
+
+if __name__ == '__main__':
+  tf.test.main()

research/object_detection/dataset_tools/tf_record_creation_util.py

@@ -20,7 +20,7 @@ from __future__ import division
 from __future__ import print_function
 
 from six.moves import range
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 
 def open_sharded_output_tfrecords(exit_stack, base_path, num_shards):

research/object_detection/dataset_tools/tf_record_creation_util_test.py

@@ -23,7 +23,7 @@ import os
 import contextlib2
 import six
 from six.moves import range
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.dataset_tools import tf_record_creation_util
 

research/object_detection/eval_util.py

@@ -24,9 +24,10 @@ import time
 
 import numpy as np
 from six.moves import range
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
+
+import tf_slim as slim
 
-from tensorflow.contrib import slim
 from object_detection.core import box_list
 from object_detection.core import box_list_ops
 from object_detection.core import keypoint_ops
@@ -553,11 +554,15 @@ def _resize_detection_masks(args):
 
 
 def _resize_groundtruth_masks(args):
-  mask, image_shape = args
+  """Resizes groundgtruth masks to the original image size."""
+  mask, true_image_shape, original_image_shape = args
+  true_height = true_image_shape[0]
+  true_width = true_image_shape[1]
+  mask = mask[:, :true_height, :true_width]
   mask = tf.expand_dims(mask, 3)
   mask = tf.image.resize_images(
       mask,
-      image_shape,
+      original_image_shape,
       method=tf.image.ResizeMethod.NEAREST_NEIGHBOR,
       align_corners=True)
   return tf.cast(tf.squeeze(mask, 3), tf.uint8)
@@ -689,7 +694,7 @@ def result_dict_for_batched_example(images,
 
   Args:
     images: A single 4D uint8 image tensor of shape [batch_size, H, W, C].
-    keys: A [batch_size] string tensor with image identifier.
+    keys: A [batch_size] string/int tensor with image identifier.
     detections: A dictionary of detections, returned from
       DetectionModel.postprocess().
     groundtruth: (Optional) Dictionary of groundtruth items, with fields:
@@ -711,6 +716,8 @@ def result_dict_for_batched_example(images,
         2] float32 tensor with keypoints (Optional).
       'groundtruth_keypoint_visibilities': [batch_size, max_number_of_boxes,
         num_keypoints] bool tensor with keypoint visibilities (Optional).
+      'groundtruth_labeled_classes': [batch_size, num_classes] int64
+        tensor of 1-indexed classes. (Optional)
     class_agnostic: Boolean indicating whether the detections are class-agnostic
       (i.e. binary). Default False.
     scale_to_absolute: Boolean indicating whether boxes and keypoints should be
@@ -762,6 +769,8 @@ def result_dict_for_batched_example(images,
       tensor with keypoints (Optional).
     'groundtruth_keypoint_visibilities': [batch_size, num_boxes, num_keypoints]
       bool tensor with keypoint visibilities (Optional).
+    'groundtruth_labeled_classes': [batch_size, num_classes]  int64 tensor
+      of 1-indexed classes. (Optional)
     'num_groundtruth_boxes': [batch_size] tensor containing the maximum number
       of groundtruth boxes per image.
 
@@ -871,7 +880,7 @@ def result_dict_for_batched_example(images,
       groundtruth[input_data_fields.groundtruth_instance_masks] = (
           shape_utils.static_or_dynamic_map_fn(
               _resize_groundtruth_masks,
-              elems=[masks, original_image_spatial_shapes],
+              elems=[masks, true_image_shapes, original_image_spatial_shapes],
               dtype=tf.uint8))
 
     output_dict.update(groundtruth)

research/object_detection/eval_util_test.py

@@ -23,7 +23,7 @@ from absl.testing import parameterized
 import numpy as np
 import six
 from six.moves import range
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection import eval_util
 from object_detection.core import standard_fields as fields

research/object_detection/export_inference_graph.py

@@ -104,7 +104,7 @@ python export_inference_graph.py \
               } \
             }"
 """
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 from google.protobuf import text_format
 from object_detection import exporter
 from object_detection.protos import pipeline_pb2

research/object_detection/export_tflite_ssd_graph.py

@@ -92,7 +92,7 @@ python object_detection/export_tflite_ssd_graph.py \
        "
 """
 
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 from google.protobuf import text_format
 from object_detection import export_tflite_ssd_graph_lib
 from object_detection.protos import pipeline_pb2