Update for py3 and some internal changes (#7786)

research/deeplab/core/dense_prediction_cell.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -25,10 +26,11 @@ from __future__ import division
 from __future__ import print_function
 
 import tensorflow as tf
+from tensorflow.contrib import slim as contrib_slim
 
 from deeplab.core import utils
 
-slim = tf.contrib.slim
+slim = contrib_slim
 
 # Local constants.
 _META_ARCHITECTURE_SCOPE = 'meta_architecture'

research/deeplab/core/dense_prediction_cell_test.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

research/deeplab/core/preprocess_utils.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -14,6 +15,11 @@
 # ==============================================================================
 
 """Utility functions related to preprocessing inputs."""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from six.moves import range
+from six.moves import zip
 import tensorflow as tf
 
 
@@ -63,7 +69,7 @@ def _image_dimensions(image, rank):
   """Returns the dimensions of an image tensor.
 
   Args:
-    image: A rank-D Tensor. For 3-D  of shape: `[height, width, channels]`.
+    image: A rank-D Tensor. For 3-D of shape: `[height, width, channels]`.
     rank: The expected rank of the image
 
   Returns:
@@ -81,6 +87,27 @@ def _image_dimensions(image, rank):
     ]
 
 
+def get_label_resize_method(label):
+  """Returns the resize method of labels depending on label dtype.
+
+  Args:
+    label: Groundtruth label tensor.
+
+  Returns:
+    tf.image.ResizeMethod.BILINEAR, if label dtype is floating.
+    tf.image.ResizeMethod.NEAREST_NEIGHBOR, if label dtype is integer.
+
+  Raises:
+    ValueError: If label is neither floating nor integer.
+  """
+  if label.dtype.is_floating:
+    return tf.image.ResizeMethod.BILINEAR
+  elif label.dtype.is_integer:
+    return tf.image.ResizeMethod.NEAREST_NEIGHBOR
+  else:
+    raise ValueError('Label type must be either floating or integer.')
+
+
 def pad_to_bounding_box(image, offset_height, offset_width, target_height,
                         target_width, pad_value):
   """Pads the given image with the given pad_value.
@@ -200,7 +227,7 @@ def _crop(image, offset_height, offset_width, crop_height, crop_width):
           tf.greater_equal(original_shape[1], crop_width)),
       ['Crop size greater than the image size.'])
 
-  offsets = tf.to_int32(tf.stack([offset_height, offset_width, 0]))
+  offsets = tf.cast(tf.stack([offset_height, offset_width, 0]), tf.int32)
 
   # Use tf.slice instead of crop_to_bounding box as it accepts tensors to
   # define the crop size.
@@ -352,10 +379,11 @@ def randomly_scale_image_and_label(image, label=None, scale=1.0):
       new_dim,
       align_corners=True), [0])
   if label is not None:
-    label = tf.squeeze(tf.image.resize_nearest_neighbor(
-        tf.expand_dims(label, 0),
+    label = tf.image.resize(
+        label,
         new_dim,
-        align_corners=True), [0])
+        method=get_label_resize_method(label),
+        align_corners=True)
 
   return image, label
 
@@ -394,6 +422,7 @@ def resize_to_range(image,
                     min_size=None,
                     max_size=None,
                     factor=None,
+                    keep_aspect_ratio=True,
                     align_corners=True,
                     label_layout_is_chw=False,
                     scope=None,
@@ -415,8 +444,12 @@ def resize_to_range(image,
     min_size: (scalar) desired size of the smaller image side.
     max_size: (scalar) maximum allowed size of the larger image side. Note
       that the output dimension is no larger than max_size and may be slightly
-      smaller than min_size when factor is not None.
+      smaller than max_size when factor is not None.
     factor: Make output size multiple of factor plus one.
+    keep_aspect_ratio: Boolean, keep aspect ratio or not. If True, the input
+      will be resized while keeping the original aspect ratio. If False, the
+      input will be resized to [max_resize_value, max_resize_value] without
+      keeping the original aspect ratio.
     align_corners: If True, exactly align all 4 corners of input and output.
     label_layout_is_chw: If true, the label has shape [channel, height, width].
       We support this case because for some instance segmentation dataset, the
@@ -441,8 +474,7 @@ def resize_to_range(image,
       # Modify the max_size to be a multiple of factor plus 1 and make sure the
       # max dimension after resizing is no larger than max_size.
       if factor is not None:
-        max_size = (max_size + (factor - (max_size - 1) % factor) % factor
-                    - factor)
+        max_size = (max_size - (max_size - 1) % factor)
 
     [orig_height, orig_width, _] = resolve_shape(image, rank=3)
     orig_height = tf.cast(orig_height, tf.float32)
@@ -451,8 +483,8 @@ def resize_to_range(image,
 
     # Calculate the larger of the possible sizes
     large_scale_factor = min_size / orig_min_size
-    large_height = tf.to_int32(tf.ceil(orig_height * large_scale_factor))
-    large_width = tf.to_int32(tf.ceil(orig_width * large_scale_factor))
+    large_height = tf.cast(tf.floor(orig_height * large_scale_factor), tf.int32)
+    large_width = tf.cast(tf.floor(orig_width * large_scale_factor), tf.int32)
     large_size = tf.stack([large_height, large_width])
 
     new_size = large_size
@@ -461,8 +493,9 @@ def resize_to_range(image,
       # is too big.
       orig_max_size = tf.maximum(orig_height, orig_width)
       small_scale_factor = max_size / orig_max_size
-      small_height = tf.to_int32(tf.ceil(orig_height * small_scale_factor))
-      small_width = tf.to_int32(tf.ceil(orig_width * small_scale_factor))
+      small_height = tf.cast(
+          tf.floor(orig_height * small_scale_factor), tf.int32)
+      small_width = tf.cast(tf.floor(orig_width * small_scale_factor), tf.int32)
       small_size = tf.stack([small_height, small_width])
       new_size = tf.cond(
           tf.cast(tf.reduce_max(large_size), tf.float32) > max_size,
@@ -471,19 +504,28 @@ def resize_to_range(image,
     # Ensure that both output sides are multiples of factor plus one.
     if factor is not None:
       new_size += (factor - (new_size - 1) % factor) % factor
-    new_tensor_list.append(tf.image.resize_images(
+    if not keep_aspect_ratio:
+      # If not keep the aspect ratio, we resize everything to max_size, allowing
+      # us to do pre-processing without extra padding.
+      new_size = [tf.reduce_max(new_size), tf.reduce_max(new_size)]
+    new_tensor_list.append(tf.image.resize(
         image, new_size, method=method, align_corners=align_corners))
     if label is not None:
       if label_layout_is_chw:
         # Input label has shape [channel, height, width].
         resized_label = tf.expand_dims(label, 3)
-        resized_label = tf.image.resize_nearest_neighbor(
-            resized_label, new_size, align_corners=align_corners)
+        resized_label = tf.image.resize(
+            resized_label,
+            new_size,
+            method=get_label_resize_method(label),
+            align_corners=align_corners)
         resized_label = tf.squeeze(resized_label, 3)
       else:
         # Input label has shape [height, width, channel].
-        resized_label = tf.image.resize_images(
-            label, new_size, method=tf.image.ResizeMethod.NEAREST_NEIGHBOR,
+        resized_label = tf.image.resize(
+            label,
+            new_size,
+            method=get_label_resize_method(label),
             align_corners=align_corners)
       new_tensor_list.append(resized_label)
     else:

research/deeplab/core/preprocess_utils_test.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -14,10 +15,14 @@
 # ==============================================================================
 
 """Tests for preprocess_utils."""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
 import numpy as np
+from six.moves import range
 import tensorflow as tf
 
-from tensorflow.python.framework import errors
 from deeplab.core import preprocess_utils
 
 
@@ -105,7 +110,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
                               [[3., 4.],
                                [5., 3.]]])
     image = tf.convert_to_tensor(numpy_image)
-    tf.set_random_seed(53)
+    tf.compat.v1.set_random_seed(53)
 
     with self.test_session() as sess:
       actual, is_flipped = preprocess_utils.flip_dim(
@@ -150,7 +155,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
     crop_height, crop_width = 10, 20
     image = np.random.randint(0, 256, size=(100, 200, 3))
 
-    tf.set_random_seed(37)
+    tf.compat.v1.set_random_seed(37)
     image_placeholder = tf.placeholder(tf.int32, shape=(None, None, 3))
     [cropped] = preprocess_utils.random_crop(
         [image_placeholder], crop_height, crop_width)
@@ -160,7 +165,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
       self.assertTupleEqual(cropped_image.shape, (crop_height, crop_width, 3))
 
   def testReturnDifferentCropAreasOnTwoEvals(self):
-    tf.set_random_seed(0)
+    tf.compat.v1.set_random_seed(0)
 
     crop_height, crop_width = 2, 3
     image = np.random.randint(0, 256, size=(100, 200, 3))
@@ -174,7 +179,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
       self.assertFalse(np.isclose(crop0, crop1).all())
 
   def testReturnConsistenCropsOfImagesInTheList(self):
-    tf.set_random_seed(0)
+    tf.compat.v1.set_random_seed(0)
 
     height, width = 10, 20
     crop_height, crop_width = 2, 3
@@ -203,7 +208,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
         [image1, image2], crop_height, crop_width)
 
     with self.test_session() as sess:
-      with self.assertRaises(errors.InvalidArgumentError):
+      with self.assertRaises(tf.errors.InvalidArgumentError):
         sess.run(cropped, feed_dict={image1: np.random.rand(4, 5, 3),
                                      image2: np.random.rand(4, 6, 1)})
 
@@ -216,7 +221,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
 
     with self.test_session() as sess:
       with self.assertRaisesWithPredicateMatch(
-          errors.InvalidArgumentError,
+          tf.errors.InvalidArgumentError,
           'Wrong height for tensor'):
         sess.run(cropped, feed_dict={image1: np.random.rand(4, 5, 3),
                                      image2: np.random.rand(3, 5, 1)})
@@ -230,7 +235,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
 
     with self.test_session() as sess:
       with self.assertRaisesWithPredicateMatch(
-          errors.InvalidArgumentError,
+          tf.errors.InvalidArgumentError,
           'Crop size greater than the image size.'):
         sess.run(cropped, feed_dict={image1: np.random.rand(4, 5, 3),
                                      image2: np.random.rand(4, 5, 1)})
@@ -284,13 +289,13 @@ class PreprocessUtilsTest(tf.test.TestCase):
       padded_image = preprocess_utils.pad_to_bounding_box(
           image_placeholder, 0, 0, 2, 1, 255)
       with self.assertRaisesWithPredicateMatch(
-          errors.InvalidArgumentError,
+          tf.errors.InvalidArgumentError,
           'target_width must be >= width'):
         padded_image.eval(feed_dict={image_placeholder: image})
       padded_image = preprocess_utils.pad_to_bounding_box(
           image_placeholder, 0, 0, 1, 2, 255)
       with self.assertRaisesWithPredicateMatch(
-          errors.InvalidArgumentError,
+          tf.errors.InvalidArgumentError,
           'target_height must be >= height'):
         padded_image.eval(feed_dict={image_placeholder: image})
 
@@ -304,7 +309,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
       padded_image = preprocess_utils.pad_to_bounding_box(
           image_placeholder, 3, 0, 4, 4, 255)
       with self.assertRaisesWithPredicateMatch(
-          errors.InvalidArgumentError,
+          tf.errors.InvalidArgumentError,
           'target size not possible with the given target offsets'):
         padded_image.eval(feed_dict={image_placeholder: image})
 
@@ -316,7 +321,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
       padded_image = preprocess_utils.pad_to_bounding_box(
           image_placeholder, 0, 0, 2, 2, 255)
       with self.assertRaisesWithPredicateMatch(
-          errors.InvalidArgumentError,
+          tf.errors.InvalidArgumentError,
           'Wrong image tensor rank'):
         padded_image.eval(feed_dict={image_placeholder: image})
 
@@ -331,7 +336,7 @@ class PreprocessUtilsTest(tf.test.TestCase):
                            (50, 100, 3),
                            (30, 100, 3)]
     for i, test_shape in enumerate(test_shapes):
-      image = tf.random_normal([test_shape[0], test_shape[1], 3])
+      image = tf.random.normal([test_shape[0], test_shape[1], 3])
       new_tensor_list = preprocess_utils.resize_to_range(
           image=image,
           label=None,
@@ -357,8 +362,8 @@ class PreprocessUtilsTest(tf.test.TestCase):
                                  (49, 97, 1),
                                  (33, 97, 1)]
     for i, test_shape in enumerate(test_shapes):
-      image = tf.random_normal([test_shape[0], test_shape[1], 3])
-      label = tf.random_normal([test_shape[0], test_shape[1], 1])
+      image = tf.random.normal([test_shape[0], test_shape[1], 3])
+      label = tf.random.normal([test_shape[0], test_shape[1], 1])
       new_tensor_list = preprocess_utils.resize_to_range(
           image=image,
           label=label,
@@ -385,8 +390,8 @@ class PreprocessUtilsTest(tf.test.TestCase):
                                  (5, 49, 97),
                                  (5, 33, 97)]
     for i, test_shape in enumerate(test_shapes):
-      image = tf.random_normal([test_shape[0], test_shape[1], 3])
-      label = tf.random_normal([5, test_shape[0], test_shape[1]])
+      image = tf.random.normal([test_shape[0], test_shape[1], 3])
+      label = tf.random.normal([5, test_shape[0], test_shape[1]])
       new_tensor_list = preprocess_utils.resize_to_range(
           image=image,
           label=label,
@@ -415,14 +420,90 @@ class PreprocessUtilsTest(tf.test.TestCase):
                                  (49, 97, 1),
                                  (33, 97, 1)]
     for i, test_shape in enumerate(test_shapes):
-      image = tf.random_normal([test_shape[0], test_shape[1], 3])
-      label = tf.random_normal([test_shape[0], test_shape[1], 1])
+      image = tf.random.normal([test_shape[0], test_shape[1], 3])
+      label = tf.random.normal([test_shape[0], test_shape[1], 1])
+      new_tensor_list = preprocess_utils.resize_to_range(
+          image=image,
+          label=label,
+          min_size=min_size,
+          max_size=max_size,
+          factor=factor,
+          align_corners=True)
+      with self.test_session() as session:
+        new_tensor_list = session.run(new_tensor_list)
+        self.assertEqual(new_tensor_list[0].shape, expected_image_shape_list[i])
+        self.assertEqual(new_tensor_list[1].shape, expected_label_shape_list[i])
+
+  def testResizeTensorsToRangeWithEqualMaxSize(self):
+    test_shapes = [[97, 38],
+                   [96, 97]]
+    # Make max_size equal to the larger value of test_shapes.
+    min_size = 97
+    max_size = 97
+    factor = 8
+    expected_image_shape_list = [(97, 41, 3),
+                                 (97, 97, 3)]
+    expected_label_shape_list = [(97, 41, 1),
+                                 (97, 97, 1)]
+    for i, test_shape in enumerate(test_shapes):
+      image = tf.random.normal([test_shape[0], test_shape[1], 3])
+      label = tf.random.normal([test_shape[0], test_shape[1], 1])
+      new_tensor_list = preprocess_utils.resize_to_range(
+          image=image,
+          label=label,
+          min_size=min_size,
+          max_size=max_size,
+          factor=factor,
+          align_corners=True)
+      with self.test_session() as session:
+        new_tensor_list = session.run(new_tensor_list)
+        self.assertEqual(new_tensor_list[0].shape, expected_image_shape_list[i])
+        self.assertEqual(new_tensor_list[1].shape, expected_label_shape_list[i])
+
+  def testResizeTensorsToRangeWithPotentialErrorInTFCeil(self):
+    test_shape = [3936, 5248]
+    # Make max_size equal to the larger value of test_shapes.
+    min_size = 1441
+    max_size = 1441
+    factor = 16
+    expected_image_shape = (1089, 1441, 3)
+    expected_label_shape = (1089, 1441, 1)
+    image = tf.random.normal([test_shape[0], test_shape[1], 3])
+    label = tf.random.normal([test_shape[0], test_shape[1], 1])
+    new_tensor_list = preprocess_utils.resize_to_range(
+        image=image,
+        label=label,
+        min_size=min_size,
+        max_size=max_size,
+        factor=factor,
+        align_corners=True)
+    with self.test_session() as session:
+      new_tensor_list = session.run(new_tensor_list)
+      self.assertEqual(new_tensor_list[0].shape, expected_image_shape)
+      self.assertEqual(new_tensor_list[1].shape, expected_label_shape)
+
+  def testResizeTensorsToRangeWithEqualMaxSizeWithoutAspectRatio(self):
+    test_shapes = [[97, 38],
+                   [96, 97]]
+    # Make max_size equal to the larger value of test_shapes.
+    min_size = 97
+    max_size = 97
+    factor = 8
+    keep_aspect_ratio = False
+    expected_image_shape_list = [(97, 97, 3),
+                                 (97, 97, 3)]
+    expected_label_shape_list = [(97, 97, 1),
+                                 (97, 97, 1)]
+    for i, test_shape in enumerate(test_shapes):
+      image = tf.random.normal([test_shape[0], test_shape[1], 3])
+      label = tf.random.normal([test_shape[0], test_shape[1], 1])
       new_tensor_list = preprocess_utils.resize_to_range(
           image=image,
           label=label,
           min_size=min_size,
           max_size=max_size,
           factor=factor,
+          keep_aspect_ratio=keep_aspect_ratio,
           align_corners=True)
       with self.test_session() as session:
         new_tensor_list = session.run(new_tensor_list)

research/deeplab/datasets/build_ade20k_data.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -15,11 +16,15 @@
 
 """Converts ADE20K data to TFRecord file format with Example protos."""
 
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
 import math
 import os
 import random
 import sys
 import build_data
+from six.moves import range
 import tensorflow as tf
 
 FLAGS = tf.app.flags.FLAGS
@@ -73,7 +78,7 @@ def _convert_dataset(dataset_split, dataset_dir, dataset_label_dir):
     seg_names.append(seg)
 
   num_images = len(img_names)
-  num_per_shard = int(math.ceil(num_images / float(_NUM_SHARDS)))
+  num_per_shard = int(math.ceil(num_images / _NUM_SHARDS))
 
   image_reader = build_data.ImageReader('jpeg', channels=3)
   label_reader = build_data.ImageReader('png', channels=1)

research/deeplab/datasets/build_cityscapes_data.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -59,12 +60,16 @@ The Example proto contains the following fields:
   image/segmentation/class/encoded: encoded semantic segmentation content.
   image/segmentation/class/format: semantic segmentation file format.
 """
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
 import glob
 import math
 import os.path
 import re
 import sys
 import build_data
+from six.moves import range
 import tensorflow as tf
 
 FLAGS = tf.app.flags.FLAGS
@@ -137,7 +142,7 @@ def _convert_dataset(dataset_split):
   label_files = _get_files('label', dataset_split)
 
   num_images = len(image_files)
-  num_per_shard = int(math.ceil(num_images / float(_NUM_SHARDS)))
+  num_per_shard = int(math.ceil(num_images / _NUM_SHARDS))
 
   image_reader = build_data.ImageReader('png', channels=3)
   label_reader = build_data.ImageReader('png', channels=1)

research/deeplab/datasets/build_voc2012_data.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -50,10 +51,14 @@ The Example proto contains the following fields:
   image/segmentation/class/encoded: encoded semantic segmentation content.
   image/segmentation/class/format: semantic segmentation file format.
 """
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
 import math
 import os.path
 import sys
 import build_data
+from six.moves import range
 import tensorflow as tf
 
 FLAGS = tf.app.flags.FLAGS
@@ -94,7 +99,7 @@ def _convert_dataset(dataset_split):
   sys.stdout.write('Processing ' + dataset)
   filenames = [x.strip('\n') for x in open(dataset_split, 'r')]
   num_images = len(filenames)
-  num_per_shard = int(math.ceil(num_images / float(_NUM_SHARDS)))
+  num_per_shard = int(math.ceil(num_images / _NUM_SHARDS))
 
   image_reader = build_data.ImageReader('jpeg', channels=3)
   label_reader = build_data.ImageReader('png', channels=1)
@@ -113,13 +118,13 @@ def _convert_dataset(dataset_split):
         # Read the image.
         image_filename = os.path.join(
             FLAGS.image_folder, filenames[i] + '.' + FLAGS.image_format)
-        image_data = tf.gfile.FastGFile(image_filename, 'rb').read()
+        image_data = tf.gfile.GFile(image_filename, 'rb').read()
         height, width = image_reader.read_image_dims(image_data)
         # Read the semantic segmentation annotation.
         seg_filename = os.path.join(
             FLAGS.semantic_segmentation_folder,
             filenames[i] + '.' + FLAGS.label_format)
-        seg_data = tf.gfile.FastGFile(seg_filename, 'rb').read()
+        seg_data = tf.gfile.GFile(seg_filename, 'rb').read()
         seg_height, seg_width = label_reader.read_image_dims(seg_data)
         if height != seg_height or width != seg_width:
           raise RuntimeError('Shape mismatched between image and label.')

research/deeplab/datasets/convert_cityscapes.sh

@@ -21,7 +21,7 @@
 # Cityscapes at https://github.com/mcordts/cityscapesScripts.
 #
 # Usage:
-#   bash ./preprocess_cityscapes.sh
+#   bash ./convert_cityscapes.sh
 #
 # The folder structure is assumed to be:
 #  + datasets

research/deeplab/datasets/data_generator.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -68,10 +69,12 @@ DatasetDescriptor = collections.namedtuple(
     ])
 
 _CITYSCAPES_INFORMATION = DatasetDescriptor(
-    splits_to_sizes={
-        'train': 2975,
-        'val': 500,
-    },
+    splits_to_sizes={'train_fine': 2975,
+                     'train_coarse': 22973,
+                     'trainval_fine': 3475,
+                     'trainval_coarse': 23473,
+                     'val_fine': 500,
+                     'test_fine': 1525},
     num_classes=19,
     ignore_label=255,
 )

research/deeplab/datasets/data_generator_test.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -14,10 +15,13 @@
 # ==============================================================================
 """Tests for deeplab.datasets.data_generator."""
 
+from __future__ import absolute_import
+from __future__ import division
 from __future__ import print_function
 
 import collections
 
+from six.moves import range
 import tensorflow as tf
 
 from deeplab import common
@@ -60,13 +64,10 @@ class DatasetTest(tf.test.TestCase):
         batch = iterator.get_next()
         batch, = sess.run([batch])
         image_attributes = _get_attributes_of_image(i)
-
-        self.assertAllEqual(batch[common.IMAGE][0], image_attributes.image)
-        self.assertAllEqual(batch[common.LABEL][0], image_attributes.label)
         self.assertEqual(batch[common.HEIGHT][0], image_attributes.height)
         self.assertEqual(batch[common.WIDTH][0], image_attributes.width)
         self.assertEqual(batch[common.IMAGE_NAME][0],
-                         image_attributes.image_name)
+                         image_attributes.image_name.encode())
 
       # All data have been read.
       with self.assertRaisesRegexp(tf.errors.OutOfRangeError, ''):
@@ -87,22 +88,22 @@ def _get_attributes_of_image(index):
   """
   if index == 0:
     return ImageAttributes(
-        image=IMAGE_1,
-        label=LABEL_1,
+        image=None,
+        label=None,
         height=366,
         width=500,
         image_name='2007_000033')
   elif index == 1:
     return ImageAttributes(
-        image=IMAGE_2,
-        label=LABEL_2,
+        image=None,
+        label=None,
         height=335,
         width=500,
         image_name='2007_000042')
   elif index == 2:
     return ImageAttributes(
-        image=IMAGE_3,
-        label=LABEL_3,
+        image=None,
+        label=None,
         height=333,
         width=500,
         image_name='2007_000061')
@@ -110,113 +111,5 @@ def _get_attributes_of_image(index):
     raise ValueError('Index can only be 0, 1 or 2.')
 
 
-IMAGE_1 = (
-    (
-        (57., 41., 18.),
-        (151.5, 138., 111.5),
-        (107., 158., 143.),
-    ),
-    (
-        (104.5, 141., 191.),
-        (101.75, 72.5, 120.75),
-        (86.5, 139.5, 120.),
-    ),
-    (
-        (96., 85., 145.),
-        (123.5, 107.5, 97.),
-        (61., 148., 116.),
-    ),
-)
-
-LABEL_1 = (
-    (
-        (70,),
-        (227,),
-        (251,),
-    ),
-    (
-        (101,),
-        (0,),
-        (10,),
-    ),
-    (
-        (145,),
-        (245,),
-        (146,),
-    ),
-)
-
-IMAGE_2 = (
-    (
-        (94., 64., 98.),
-        (145.5, 136.5, 134.5),
-        (108., 162., 172.),
-    ),
-    (
-        (168., 157., 213.),
-        (161.5, 154.5, 148.),
-        (25., 46., 93.),
-    ),
-    (
-        (255., 204., 237.),
-        (124., 102., 126.5),
-        (155., 181., 82.),
-    ),
-)
-
-LABEL_2 = (
-    (
-        (44,),
-        (146,),
-        (121,),
-    ),
-    (
-        (108,),
-        (118,),
-        (6,),
-    ),
-    (
-        (246,),
-        (121,),
-        (108,),
-    ),
-)
-
-IMAGE_3 = (
-    (
-        (235., 173., 150.),
-        (145.5, 83.5, 102.),
-        (82., 149., 158.),
-    ),
-    (
-        (130., 95., 14.),
-        (132.5, 141.5, 93.),
-        (119., 85., 86.),
-    ),
-    (
-        (127.5, 127.5, 127.5),
-        (127.5, 127.5, 127.5),
-        (127.5, 127.5, 127.5),
-    ),
-)
-
-LABEL_3 = (
-    (
-        (91,),
-        (120,),
-        (132,),
-    ),
-    (
-        (135,),
-        (139,),
-        (72,),
-    ),
-    (
-        (255,),
-        (255,),
-        (255,),
-    ),
-)
-
 if __name__ == '__main__':
   tf.test.main()

research/deeplab/deeplab_demo.ipynb

@@ -291,7 +291,7 @@
         "url for inference.\n",
         "\n",
         "Note that this colab uses single scale inference for fast computation,\n",
-        "so the results may slightly differ from the visualizations in the \n",
+        "so the results may slightly differ from the visualizations in the\n",
         "[README](https://github.com/tensorflow/models/blob/master/research/deeplab/README.md) file,\n",
         "which uses multi-scale and left-right flipped inputs."
       ]

research/deeplab/deprecated/segmentation_dataset.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -51,8 +52,9 @@ References:
 import collections
 import os.path
 import tensorflow as tf
+from tensorflow.contrib import slim as contrib_slim
 
-slim = tf.contrib.slim
+slim = contrib_slim
 
 dataset = slim.dataset
 

research/deeplab/eval.py

+# Lint as: python2, python3
 # Copyright 2018 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -17,7 +18,13 @@
 See model.py for more details and usage.
 """
 
+import numpy as np
+import six
 import tensorflow as tf
+from tensorflow.contrib import metrics as contrib_metrics
+from tensorflow.contrib import quantize as contrib_quantize
+from tensorflow.contrib import tfprof as contrib_tfprof
+from tensorflow.contrib import training as contrib_training
 from deeplab import common
 from deeplab import model
 from deeplab.datasets import data_generator
@@ -149,12 +156,44 @@ def main(unused_argv):
       predictions_tag += '_flipped'
 
     # Define the evaluation metric.
-    miou, update_op = tf.metrics.mean_iou(
-        predictions, labels, dataset.num_of_classes, weights=weights)
-    tf.summary.scalar(predictions_tag, miou)
-
-    summary_op = tf.summary.merge_all()
-    summary_hook = tf.contrib.training.SummaryAtEndHook(
+    metric_map = {}
+    num_classes = dataset.num_of_classes
+    metric_map['eval/%s_overall' % predictions_tag] = tf.metrics.mean_iou(
+        labels=labels, predictions=predictions, num_classes=num_classes,
+        weights=weights)
+    # IoU for each class.
+    one_hot_predictions = tf.one_hot(predictions, num_classes)
+    one_hot_predictions = tf.reshape(one_hot_predictions, [-1, num_classes])
+    one_hot_labels = tf.one_hot(labels, num_classes)
+    one_hot_labels = tf.reshape(one_hot_labels, [-1, num_classes])
+    for c in range(num_classes):
+      predictions_tag_c = '%s_class_%d' % (predictions_tag, c)
+      tp, tp_op = tf.metrics.true_positives(
+          labels=one_hot_labels[:, c], predictions=one_hot_predictions[:, c],
+          weights=weights)
+      fp, fp_op = tf.metrics.false_positives(
+          labels=one_hot_labels[:, c], predictions=one_hot_predictions[:, c],
+          weights=weights)
+      fn, fn_op = tf.metrics.false_negatives(
+          labels=one_hot_labels[:, c], predictions=one_hot_predictions[:, c],
+          weights=weights)
+      tp_fp_fn_op = tf.group(tp_op, fp_op, fn_op)
+      iou = tf.where(tf.greater(tp + fn, 0.0),
+                     tp / (tp + fn + fp),
+                     tf.constant(np.NaN))
+      metric_map['eval/%s' % predictions_tag_c] = (iou, tp_fp_fn_op)
+
+    (metrics_to_values,
+     metrics_to_updates) = contrib_metrics.aggregate_metric_map(metric_map)
+
+    summary_ops = []
+    for metric_name, metric_value in six.iteritems(metrics_to_values):
+      op = tf.summary.scalar(metric_name, metric_value)
+      op = tf.Print(op, [metric_value], metric_name)
+      summary_ops.append(op)
+
+    summary_op = tf.summary.merge(summary_ops)
+    summary_hook = contrib_training.SummaryAtEndHook(
         log_dir=FLAGS.eval_logdir, summary_op=summary_op)
     hooks = [summary_hook]
 
@@ -163,19 +202,19 @@ def main(unused_argv):
       num_eval_iters = FLAGS.max_number_of_evaluations
 
     if FLAGS.quantize_delay_step >= 0:
-      tf.contrib.quantize.create_eval_graph()
+      contrib_quantize.create_eval_graph()
 
-    tf.contrib.tfprof.model_analyzer.print_model_analysis(
+    contrib_tfprof.model_analyzer.print_model_analysis(
         tf.get_default_graph(),
-        tfprof_options=tf.contrib.tfprof.model_analyzer.
-        TRAINABLE_VARS_PARAMS_STAT_OPTIONS)
-    tf.contrib.tfprof.model_analyzer.print_model_analysis(
+        tfprof_options=contrib_tfprof.model_analyzer
+        .TRAINABLE_VARS_PARAMS_STAT_OPTIONS)
+    contrib_tfprof.model_analyzer.print_model_analysis(
         tf.get_default_graph(),
-        tfprof_options=tf.contrib.tfprof.model_analyzer.FLOAT_OPS_OPTIONS)
-    tf.contrib.training.evaluate_repeatedly(
-        master=FLAGS.master,
+        tfprof_options=contrib_tfprof.model_analyzer.FLOAT_OPS_OPTIONS)
+    contrib_training.evaluate_repeatedly(
         checkpoint_dir=FLAGS.checkpoint_dir,
-        eval_ops=[update_op],
+        master=FLAGS.master,
+        eval_ops=list(metrics_to_updates.values()),
         max_number_of_evaluations=num_eval_iters,
         hooks=hooks,
         eval_interval_secs=FLAGS.eval_interval_secs)

research/deeplab/evaluation/eval_coco_format.py

+# Lint as: python2, python3
 # Copyright 2019 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

research/deeplab/evaluation/eval_coco_format_test.py

+# Lint as: python2, python3
 # Copyright 2019 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -52,7 +53,8 @@ class EvalCocoFormatTest(absltest.TestCase):
         ignored_label=0,
         max_instances_per_category=256,
         intersection_offset=(256 * 256))
-    self.assertCountEqual(deeplab_results.keys(), ['All', 'Things', 'Stuff'])
+    self.assertCountEqual(
+        list(deeplab_results.keys()), ['All', 'Things', 'Stuff'])
     for cat_group in ['All', 'Things', 'Stuff']:
       self.assertCountEqual(deeplab_results[cat_group], ['pq', 'sq', 'rq', 'n'])
       for metric in ['pq', 'sq', 'rq', 'n']:
@@ -77,7 +79,8 @@ class EvalCocoFormatTest(absltest.TestCase):
         max_instances_per_category=256,
         intersection_offset=(256 * 256),
         normalize_by_image_size=False)
-    self.assertCountEqual(deeplab_results.keys(), ['All', 'Things', 'Stuff'])
+    self.assertCountEqual(
+        list(deeplab_results.keys()), ['All', 'Things', 'Stuff'])
     for cat_group in ['All', 'Things', 'Stuff']:
       self.assertCountEqual(deeplab_results[cat_group], ['pc', 'n'])
     self.assertAlmostEqual(deeplab_results['All']['pc'], 0.68210561)
@@ -105,7 +108,8 @@ class EvalCocoFormatTest(absltest.TestCase):
         max_instances_per_category=256,
         intersection_offset=(256 * 256),
         normalize_by_image_size=True)
-    self.assertCountEqual(deeplab_results.keys(), ['All', 'Things', 'Stuff'])
+    self.assertCountEqual(
+        list(deeplab_results.keys()), ['All', 'Things', 'Stuff'])
     self.assertAlmostEqual(deeplab_results['All']['pc'], 0.68214908840)
 
   def test_pc_with_multiple_workers(self):
@@ -127,7 +131,8 @@ class EvalCocoFormatTest(absltest.TestCase):
         intersection_offset=(256 * 256),
         num_workers=3,
         normalize_by_image_size=False)
-    self.assertCountEqual(deeplab_results.keys(), ['All', 'Things', 'Stuff'])
+    self.assertCountEqual(
+        list(deeplab_results.keys()), ['All', 'Things', 'Stuff'])
     self.assertAlmostEqual(deeplab_results['All']['pc'], 0.68210561668)
 
 

research/deeplab/evaluation/panoptic_quality.py

+# Lint as: python2, python3
 # Copyright 2019 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

research/deeplab/evaluation/panoptic_quality_test.py

+# Lint as: python2, python3
 # Copyright 2019 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

research/deeplab/evaluation/parsing_covering.py

+# Lint as: python2, python3
 # Copyright 2019 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

research/deeplab/evaluation/parsing_covering_test.py

+# Lint as: python2, python3
 # Copyright 2019 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

research/deeplab/evaluation/streaming_metrics.py

+# Lint as: python2, python3
 # Copyright 2019 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");