diff --git a/tensorflow/python/feature_column/feature_column_v2_test.py b/tensorflow/python/feature_column/feature_column_v2_test.py
index 85c21c0d38e..a72bdf1f631 100644
--- a/tensorflow/python/feature_column/feature_column_v2_test.py
+++ b/tensorflow/python/feature_column/feature_column_v2_test.py
@@ -27,7 +27,6 @@ from tensorflow.core.example import example_pb2
 from tensorflow.core.example import feature_pb2
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.core.protobuf import rewriter_config_pb2
-from tensorflow.python import keras
 from tensorflow.python.client import session
 from tensorflow.python.eager import backprop
 from tensorflow.python.eager import context
@@ -41,6 +40,7 @@ from tensorflow.python.framework import errors
 from tensorflow.python.framework import ops
 from tensorflow.python.framework import sparse_tensor
 from tensorflow.python.framework import test_util
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import init_ops
 from tensorflow.python.ops import lookup_ops
@@ -2085,7 +2085,7 @@ class LinearModelTest(test.TestCase):
 
     x = {'a': np.random.random((10, 1))}
     y = np.random.randint(20, size=(10, 1))
-    y = keras.utils.to_categorical(y, num_classes=20)
+    y = np_utils.to_categorical(y, num_classes=20)
     model.fit(x, y, epochs=1, batch_size=5)
     model.fit(x, y, epochs=1, batch_size=5)
     model.evaluate(x, y, batch_size=5)
diff --git a/tensorflow/python/feature_column/sequence_feature_column.py b/tensorflow/python/feature_column/sequence_feature_column.py
index 8dce0926a23..0afe64d6de9 100644
--- a/tensorflow/python/feature_column/sequence_feature_column.py
+++ b/tensorflow/python/feature_column/sequence_feature_column.py
@@ -30,7 +30,7 @@ from tensorflow.python.feature_column import utils as fc_utils
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
 from tensorflow.python.framework import tensor_shape
-from tensorflow.python.keras import utils
+from tensorflow.python.keras.utils import generic_utils
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import check_ops
 from tensorflow.python.ops import parsing_ops
@@ -585,7 +585,8 @@ class SequenceNumericColumn(
   def get_config(self):
     """See 'FeatureColumn` base class."""
     config = dict(zip(self._fields, self))
-    config['normalizer_fn'] = utils.serialize_keras_object(self.normalizer_fn)
+    config['normalizer_fn'] = generic_utils.serialize_keras_object(
+        self.normalizer_fn)
     config['dtype'] = self.dtype.name
     return config
 
@@ -594,7 +595,7 @@ class SequenceNumericColumn(
     """See 'FeatureColumn` base class."""
     fc._check_config_keys(config, cls._fields)
     kwargs = fc._standardize_and_copy_config(config)
-    kwargs['normalizer_fn'] = utils.deserialize_keras_object(
+    kwargs['normalizer_fn'] = generic_utils.deserialize_keras_object(
         config['normalizer_fn'], custom_objects=custom_objects)
     kwargs['dtype'] = dtypes.as_dtype(config['dtype'])
     return cls(**kwargs)
diff --git a/tensorflow/python/feature_column/serialization.py b/tensorflow/python/feature_column/serialization.py
index 81058d7b657..eb183d7bbcd 100644
--- a/tensorflow/python/feature_column/serialization.py
+++ b/tensorflow/python/feature_column/serialization.py
@@ -28,7 +28,7 @@ from tensorflow.python.util.lazy_loader import LazyLoader
 # Prevent circular dependencies with Keras serialization.
 generic_utils = LazyLoader(
     'generic_utils', globals(),
-    'tensorflow.python.keras.utils')
+    'tensorflow.python.keras.utils.generic_utils')
 
 _FEATURE_COLUMNS = [
     fc_lib.BucketizedColumn, fc_lib.CrossedColumn, fc_lib.EmbeddingColumn,
diff --git a/tensorflow/python/keras/BUILD b/tensorflow/python/keras/BUILD
index faef05da80b..9371bb73764 100755
--- a/tensorflow/python/keras/BUILD
+++ b/tensorflow/python/keras/BUILD
@@ -51,6 +51,7 @@ py_library(
         "preprocessing/text.py",
         "testing_utils.py",
         "utils/__init__.py",
+        "utils/all_utils.py",
         "utils/multi_gpu_utils.py",
         "utils/np_utils.py",
         "utils/vis_utils.py",
diff --git a/tensorflow/python/keras/__init__.py b/tensorflow/python/keras/__init__.py
index 7321876b2ad..f7cab0e7b18 100644
--- a/tensorflow/python/keras/__init__.py
+++ b/tensorflow/python/keras/__init__.py
@@ -41,7 +41,6 @@ from tensorflow.python.keras import optimizers
 from tensorflow.python.keras import premade
 from tensorflow.python.keras import preprocessing
 from tensorflow.python.keras import regularizers
-from tensorflow.python.keras import utils
 from tensorflow.python.keras.layers import Input
 from tensorflow.python.keras.models import Model
 from tensorflow.python.keras.models import Sequential
diff --git a/tensorflow/python/keras/api/BUILD b/tensorflow/python/keras/api/BUILD
index 879aeabdc24..dc162b9fc44 100644
--- a/tensorflow/python/keras/api/BUILD
+++ b/tensorflow/python/keras/api/BUILD
@@ -13,6 +13,13 @@ package(
 keras_packages = [
     "tensorflow.python",
     "tensorflow.python.keras",
+    "tensorflow.python.keras.utils.data_utils",
+    "tensorflow.python.keras.utils.generic_utils",
+    "tensorflow.python.keras.utils.io_utils",
+    "tensorflow.python.keras.utils.layer_utils",
+    "tensorflow.python.keras.utils.multi_gpu_utils",
+    "tensorflow.python.keras.utils.np_utils",
+    "tensorflow.python.keras.utils.vis_utils",
     "tensorflow.python.keras.wrappers.scikit_learn",
 ]
 
diff --git a/tensorflow/python/keras/applications/__init__.py b/tensorflow/python/keras/applications/__init__.py
index 6e343f6f5af..ef89793fd73 100644
--- a/tensorflow/python/keras/applications/__init__.py
+++ b/tensorflow/python/keras/applications/__init__.py
@@ -25,7 +25,7 @@ from tensorflow.python.keras import backend
 from tensorflow.python.keras import engine
 from tensorflow.python.keras import layers
 from tensorflow.python.keras import models
-from tensorflow.python.keras import utils
+from tensorflow.python.keras.utils import all_utils
 from tensorflow.python.util import tf_inspect
 
 
@@ -45,7 +45,7 @@ def keras_modules_injection(base_fun):
     if 'layers' not in kwargs:
       kwargs['layers'] = layers
     kwargs['models'] = models
-    kwargs['utils'] = utils
+    kwargs['utils'] = all_utils
     return base_fun(*args, **kwargs)
   return wrapper
 
diff --git a/tensorflow/python/keras/callbacks_test.py b/tensorflow/python/keras/callbacks_test.py
index 2c556dead3d..9346377d0bd 100644
--- a/tensorflow/python/keras/callbacks_test.py
+++ b/tensorflow/python/keras/callbacks_test.py
@@ -41,6 +41,8 @@ from tensorflow.python.keras import testing_utils
 from tensorflow.python.keras.engine import base_layer
 from tensorflow.python.keras.engine import sequential
 from tensorflow.python.keras.optimizer_v2 import gradient_descent
+from tensorflow.python.keras.optimizer_v2 import learning_rate_schedule
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import math_ops
 from tensorflow.python.ops import summary_ops_v2
@@ -49,7 +51,6 @@ from tensorflow.python.platform import tf_logging as logging
 from tensorflow.python.summary import summary_iterator
 from tensorflow.python.training import adam
 from tensorflow.python.training import checkpoint_management
-from tensorflow.python.keras.optimizer_v2 import learning_rate_schedule
 
 try:
   import h5py  # pylint:disable=g-import-not-at-top
@@ -371,8 +372,8 @@ class KerasCallbacksTest(keras_parameterized.TestCase):
         test_samples=TEST_SAMPLES,
         input_shape=(INPUT_DIM,),
         num_classes=NUM_CLASSES)
-    y_test = keras.utils.to_categorical(y_test)
-    y_train = keras.utils.to_categorical(y_train)
+    y_test = np_utils.to_categorical(y_test)
+    y_train = np_utils.to_categorical(y_train)
     # case 1
     monitor = 'val_loss'
     save_best_only = False
@@ -803,8 +804,8 @@ class KerasCallbacksTest(keras_parameterized.TestCase):
           test_samples=TEST_SAMPLES,
           input_shape=(INPUT_DIM,),
           num_classes=NUM_CLASSES)
-      y_test = keras.utils.to_categorical(y_test)
-      y_train = keras.utils.to_categorical(y_train)
+      y_test = np_utils.to_categorical(y_test)
+      y_train = np_utils.to_categorical(y_train)
       model = testing_utils.get_small_sequential_mlp(
           num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
       model.compile(
@@ -932,8 +933,8 @@ class KerasCallbacksTest(keras_parameterized.TestCase):
           test_samples=TEST_SAMPLES,
           input_shape=(INPUT_DIM,),
           num_classes=NUM_CLASSES)
-      y_test = keras.utils.to_categorical(y_test)
-      y_train = keras.utils.to_categorical(y_train)
+      y_test = np_utils.to_categorical(y_test)
+      y_train = np_utils.to_categorical(y_train)
       model = testing_utils.get_small_sequential_mlp(
           num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
       model.compile(
@@ -1003,8 +1004,8 @@ class KerasCallbacksTest(keras_parameterized.TestCase):
           test_samples=TEST_SAMPLES,
           input_shape=(INPUT_DIM,),
           num_classes=NUM_CLASSES)
-      y_test = keras.utils.to_categorical(y_test)
-      y_train = keras.utils.to_categorical(y_train)
+      y_test = np_utils.to_categorical(y_test)
+      y_train = np_utils.to_categorical(y_train)
 
       def make_model():
         random_seed.set_random_seed(1234)
@@ -1110,8 +1111,8 @@ class KerasCallbacksTest(keras_parameterized.TestCase):
           test_samples=TEST_SAMPLES,
           input_shape=(INPUT_DIM,),
           num_classes=NUM_CLASSES)
-      y_test = keras.utils.to_categorical(y_test)
-      y_train = keras.utils.to_categorical(y_train)
+      y_test = np_utils.to_categorical(y_test)
+      y_train = np_utils.to_categorical(y_train)
 
       def make_model():
         np.random.seed(1337)
@@ -1189,8 +1190,8 @@ class KerasCallbacksTest(keras_parameterized.TestCase):
           input_shape=(INPUT_DIM,),
           num_classes=NUM_CLASSES)
 
-      y_test = keras.utils.to_categorical(y_test)
-      y_train = keras.utils.to_categorical(y_train)
+      y_test = np_utils.to_categorical(y_test)
+      y_train = np_utils.to_categorical(y_train)
       cbks = [keras.callbacks.TerminateOnNaN(), keras.callbacks.CSVLogger(fp)]
       model = keras.models.Sequential()
       for _ in range(5):
@@ -1241,8 +1242,8 @@ class KerasCallbacksTest(keras_parameterized.TestCase):
           input_shape=(INPUT_DIM,),
           num_classes=NUM_CLASSES)
 
-      y_test = keras.utils.to_categorical(y_test)
-      y_train = keras.utils.to_categorical(y_train)
+      y_test = np_utils.to_categorical(y_test)
+      y_train = np_utils.to_categorical(y_train)
       cbks = [keras.callbacks.TerminateOnNaN()]
       model = keras.models.Sequential()
       initializer = keras.initializers.Constant(value=1e5)
@@ -1278,8 +1279,8 @@ class KerasCallbacksTest(keras_parameterized.TestCase):
           test_samples=TEST_SAMPLES,
           input_shape=(INPUT_DIM,),
           num_classes=NUM_CLASSES)
-      y_test = keras.utils.to_categorical(y_test)
-      y_train = keras.utils.to_categorical(y_train)
+      y_test = np_utils.to_categorical(y_test)
+      y_train = np_utils.to_categorical(y_train)
       model = keras.models.Sequential()
       model.add(
           keras.layers.Dense(
diff --git a/tensorflow/python/keras/callbacks_v1_test.py b/tensorflow/python/keras/callbacks_v1_test.py
index 7cb02b84c8f..7df695a1a57 100644
--- a/tensorflow/python/keras/callbacks_v1_test.py
+++ b/tensorflow/python/keras/callbacks_v1_test.py
@@ -29,6 +29,7 @@ from tensorflow.python import keras
 from tensorflow.python.framework import test_util
 from tensorflow.python.keras import callbacks_v1
 from tensorflow.python.keras import testing_utils
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.platform import test
 from tensorflow.python.training import adam
 
@@ -55,8 +56,8 @@ class TestTensorBoardV1(test.TestCase):
         test_samples=TEST_SAMPLES,
         input_shape=(INPUT_DIM,),
         num_classes=NUM_CLASSES)
-    y_test = keras.utils.to_categorical(y_test)
-    y_train = keras.utils.to_categorical(y_train)
+    y_test = np_utils.to_categorical(y_test)
+    y_train = np_utils.to_categorical(y_train)
 
     def data_generator(train):
       if train:
@@ -164,8 +165,8 @@ class TestTensorBoardV1(test.TestCase):
           test_samples=TEST_SAMPLES,
           input_shape=(INPUT_DIM,),
           num_classes=NUM_CLASSES)
-      y_test = keras.utils.to_categorical(y_test)
-      y_train = keras.utils.to_categorical(y_train)
+      y_test = np_utils.to_categorical(y_test)
+      y_train = np_utils.to_categorical(y_train)
 
       def data_generator(train):
         if train:
@@ -268,8 +269,8 @@ class TestTensorBoardV1(test.TestCase):
         test_samples=TEST_SAMPLES,
         input_shape=(INPUT_DIM,),
         num_classes=NUM_CLASSES)
-    y_test = keras.utils.to_categorical(y_test)
-    y_train = keras.utils.to_categorical(y_train)
+    y_test = np_utils.to_categorical(y_test)
+    y_train = np_utils.to_categorical(y_train)
 
     with self.cached_session():
       model = keras.models.Sequential()
@@ -364,8 +365,8 @@ class TestTensorBoardV1(test.TestCase):
           test_samples=TEST_SAMPLES,
           input_shape=(INPUT_DIM,),
           num_classes=NUM_CLASSES)
-      y_test = keras.utils.to_categorical(y_test)
-      y_train = keras.utils.to_categorical(y_train)
+      y_test = np_utils.to_categorical(y_test)
+      y_train = np_utils.to_categorical(y_train)
 
       model = testing_utils.get_small_sequential_mlp(
           num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
@@ -475,8 +476,8 @@ class TestTensorBoardV1(test.TestCase):
         test_samples=TEST_SAMPLES,
         input_shape=(INPUT_DIM,),
         num_classes=NUM_CLASSES)
-    y_test = keras.utils.to_categorical(y_test)
-    y_train = keras.utils.to_categorical(y_train)
+    y_test = np_utils.to_categorical(y_test)
+    y_train = np_utils.to_categorical(y_train)
 
     model = testing_utils.get_small_sequential_mlp(
         num_hidden=NUM_HIDDEN, num_classes=NUM_CLASSES, input_dim=INPUT_DIM)
diff --git a/tensorflow/python/keras/distribute/distribute_strategy_test.py b/tensorflow/python/keras/distribute/distribute_strategy_test.py
index 182f0ef9490..aa92886e09e 100644
--- a/tensorflow/python/keras/distribute/distribute_strategy_test.py
+++ b/tensorflow/python/keras/distribute/distribute_strategy_test.py
@@ -37,6 +37,7 @@ from tensorflow.python.keras.distribute import distributed_training_utils
 from tensorflow.python.keras.engine import base_layer_utils
 from tensorflow.python.keras.mixed_precision.experimental import policy
 from tensorflow.python.keras.optimizer_v2 import gradient_descent as gradient_descent_keras
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import check_ops
 from tensorflow.python.ops import math_ops
@@ -112,10 +113,10 @@ def get_multi_inputs_multi_outputs_data():
       num_classes=2,
       random_seed=_RANDOM_SEED)
 
-  c_train = keras.utils.to_categorical(c_train)
-  c_test = keras.utils.to_categorical(c_test)
-  d_train = keras.utils.to_categorical(d_train)
-  d_test = keras.utils.to_categorical(d_test)
+  c_train = np_utils.to_categorical(c_train)
+  c_test = np_utils.to_categorical(c_test)
+  d_train = np_utils.to_categorical(d_train)
+  d_test = np_utils.to_categorical(d_test)
 
   train_data = {
       'input_a': a_train,
diff --git a/tensorflow/python/keras/engine/feature_columns_integration_test.py b/tensorflow/python/keras/engine/feature_columns_integration_test.py
index f50508c11ee..6ab97023729 100644
--- a/tensorflow/python/keras/engine/feature_columns_integration_test.py
+++ b/tensorflow/python/keras/engine/feature_columns_integration_test.py
@@ -26,6 +26,7 @@ from tensorflow.python.feature_column import feature_column_lib as fc
 from tensorflow.python.keras import keras_parameterized
 from tensorflow.python.keras import metrics as metrics_module
 from tensorflow.python.keras import testing_utils
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.platform import test
 
 
@@ -64,7 +65,7 @@ class FeatureColumnsIntegrationTest(keras_parameterized.TestCase):
 
     x = {'a': np.random.random((10, 1))}
     y = np.random.randint(20, size=(10, 1))
-    y = keras.utils.to_categorical(y, num_classes=20)
+    y = np_utils.to_categorical(y, num_classes=20)
     model.fit(x, y, epochs=1, batch_size=5)
     model.fit(x, y, epochs=1, batch_size=5)
     model.evaluate(x, y, batch_size=5)
@@ -86,7 +87,7 @@ class FeatureColumnsIntegrationTest(keras_parameterized.TestCase):
         experimental_run_tf_function=testing_utils.should_run_tf_function())
 
     y = np.random.randint(20, size=(100, 1))
-    y = keras.utils.to_categorical(y, num_classes=20)
+    y = np_utils.to_categorical(y, num_classes=20)
     x = {'a': np.random.random((100, 1))}
     ds1 = dataset_ops.Dataset.from_tensor_slices(x)
     ds2 = dataset_ops.Dataset.from_tensor_slices(y)
@@ -151,7 +152,7 @@ class FeatureColumnsIntegrationTest(keras_parameterized.TestCase):
 
     x = {'a': np.random.random((10, 1)), 'b': np.random.random((10, 1))}
     y = np.random.randint(20, size=(10, 1))
-    y = keras.utils.to_categorical(y, num_classes=20)
+    y = np_utils.to_categorical(y, num_classes=20)
     dnn_model.fit(x=x, y=y, epochs=1, batch_size=5)
     dnn_model.fit(x=x, y=y, epochs=1, batch_size=5)
     dnn_model.evaluate(x=x, y=y, batch_size=5)
@@ -172,7 +173,7 @@ class FeatureColumnsIntegrationTest(keras_parameterized.TestCase):
         experimental_run_tf_function=testing_utils.should_run_tf_function())
 
     y = np.random.randint(20, size=(100, 1))
-    y = keras.utils.to_categorical(y, num_classes=20)
+    y = np_utils.to_categorical(y, num_classes=20)
     x = {'a': np.random.random((100, 1)), 'b': np.random.random((100, 1))}
     ds1 = dataset_ops.Dataset.from_tensor_slices(x)
     ds2 = dataset_ops.Dataset.from_tensor_slices(y)
diff --git a/tensorflow/python/keras/engine/training_generator_test.py b/tensorflow/python/keras/engine/training_generator_test.py
index cebc1ba5e5d..30e59114e75 100644
--- a/tensorflow/python/keras/engine/training_generator_test.py
+++ b/tensorflow/python/keras/engine/training_generator_test.py
@@ -416,7 +416,7 @@ class TestGeneratorMethodsWithSequences(keras_parameterized.TestCase):
   @data_utils.dont_use_multiprocessing_pool
   def test_training_with_sequences(self):
 
-    class DummySequence(keras.utils.Sequence):
+    class DummySequence(data_utils.Sequence):
 
       def __getitem__(self, idx):
         return np.zeros([10, 2]), np.ones([10, 4])
@@ -449,7 +449,7 @@ class TestGeneratorMethodsWithSequences(keras_parameterized.TestCase):
   def test_sequence_input_to_fit_eval_predict(self):
     val_data = np.ones([10, 10], np.float32), np.ones([10, 1], np.float32)
 
-    class CustomSequence(keras.utils.Sequence):
+    class CustomSequence(data_utils.Sequence):
 
       def __getitem__(self, idx):
         return np.ones([10, 10], np.float32), np.ones([10, 1], np.float32)
@@ -457,7 +457,7 @@ class TestGeneratorMethodsWithSequences(keras_parameterized.TestCase):
       def __len__(self):
         return 2
 
-    class CustomSequenceChangingBatchSize(keras.utils.Sequence):
+    class CustomSequenceChangingBatchSize(data_utils.Sequence):
 
       def __getitem__(self, idx):
         batch_size = 10 - idx
diff --git a/tensorflow/python/keras/engine/training_test.py b/tensorflow/python/keras/engine/training_test.py
index 81eb52471d1..722a2303f9c 100644
--- a/tensorflow/python/keras/engine/training_test.py
+++ b/tensorflow/python/keras/engine/training_test.py
@@ -39,9 +39,11 @@ from tensorflow.python.keras import keras_parameterized
 from tensorflow.python.keras import losses
 from tensorflow.python.keras import metrics as metrics_module
 from tensorflow.python.keras import testing_utils
-from tensorflow.python.keras.engine import training_utils
 from tensorflow.python.keras.callbacks import Callback
+from tensorflow.python.keras.engine import training_utils
 from tensorflow.python.keras.optimizer_v2 import gradient_descent
+from tensorflow.python.keras.utils import data_utils
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import math_ops
 from tensorflow.python.ops import sparse_ops
@@ -1204,8 +1206,8 @@ class TrainingTest(keras_parameterized.TestCase):
     with context.eager_mode():
       np.random.seed(1337)
       train_x = np.ones((100, 4))
-      train_y = keras.utils.to_categorical(np.random.randint(0, 1,
-                                                             size=(100, 1)), 2)
+      train_y = np_utils.to_categorical(
+          np.random.randint(0, 1, size=(100, 1)), 2)
 
       reference_model = testing_utils.get_small_sequential_mlp(16, 2,
                                                                input_dim=4)
@@ -1599,7 +1601,7 @@ class TestExceptionsAndWarnings(keras_parameterized.TestCase):
   @keras_parameterized.run_all_keras_modes
   def test_invalid_batch_size_argument_with_sequence_input(self):
 
-    class DummySequence(keras.utils.Sequence):
+    class DummySequence(data_utils.Sequence):
 
       def __getitem__(self, idx):
         return np.zeros([10, 2]), np.ones([10, 4])
@@ -1673,8 +1675,8 @@ class LossWeightingTest(keras_parameterized.TestCase):
     int_y_test = y_test.copy()
     int_y_train = y_train.copy()
     # convert class vectors to binary class matrices
-    y_train = keras.utils.to_categorical(y_train, num_classes)
-    y_test = keras.utils.to_categorical(y_test, num_classes)
+    y_train = np_utils.to_categorical(y_train, num_classes)
+    y_test = np_utils.to_categorical(y_test, num_classes)
     test_ids = np.where(int_y_test == np.array(weighted_class))[0]
 
     class_weight = dict([(i, 1.) for i in range(num_classes)])
@@ -1742,8 +1744,8 @@ class LossWeightingTest(keras_parameterized.TestCase):
     int_y_test = y_test.copy()
     int_y_train = y_train.copy()
     # convert class vectors to binary class matrices
-    y_train = keras.utils.to_categorical(y_train, num_classes)
-    y_test = keras.utils.to_categorical(y_test, num_classes)
+    y_train = np_utils.to_categorical(y_train, num_classes)
+    y_test = np_utils.to_categorical(y_test, num_classes)
     test_ids = np.where(int_y_test == np.array(weighted_class))[0]
 
     sample_weight = np.ones((y_train.shape[0]))
@@ -1812,8 +1814,8 @@ class LossWeightingTest(keras_parameterized.TestCase):
       int_y_test = y_test.copy()
       int_y_train = y_train.copy()
       # convert class vectors to binary class matrices
-      y_train = keras.utils.to_categorical(y_train, num_classes)
-      y_test = keras.utils.to_categorical(y_test, num_classes)
+      y_train = np_utils.to_categorical(y_train, num_classes)
+      y_test = np_utils.to_categorical(y_test, num_classes)
       test_ids = np.where(int_y_test == np.array(weighted_class))[0]
 
       sample_weight = np.ones((y_train.shape[0]))
@@ -1938,7 +1940,7 @@ class LossWeightingTest(keras_parameterized.TestCase):
           input_shape=(input_dim,),
           num_classes=num_classes)
       # convert class vectors to binary class matrices
-      y_train = keras.utils.to_categorical(y_train, num_classes)
+      y_train = np_utils.to_categorical(y_train, num_classes)
       class_weight = dict([(i, 1.) for i in range(num_classes)])
 
       del class_weight[1]
diff --git a/tensorflow/python/keras/integration_test.py b/tensorflow/python/keras/integration_test.py
index 8510db99e38..747c5043d04 100644
--- a/tensorflow/python/keras/integration_test.py
+++ b/tensorflow/python/keras/integration_test.py
@@ -28,6 +28,7 @@ from tensorflow.python.eager import context
 from tensorflow.python.framework import dtypes
 from tensorflow.python.keras import keras_parameterized
 from tensorflow.python.keras import testing_utils
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.ops import nn_ops as nn
 from tensorflow.python.ops import rnn_cell
 from tensorflow.python.platform import test
@@ -62,7 +63,7 @@ class VectorClassificationIntegrationTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(10,),
         num_classes=2)
-    y_train = keras.utils.to_categorical(y_train)
+    y_train = np_utils.to_categorical(y_train)
 
     model = testing_utils.get_model_from_layers(
         [keras.layers.Dense(16, activation='relu'),
@@ -93,7 +94,7 @@ class VectorClassificationIntegrationTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(10,),
         num_classes=2)
-    y_train = keras.utils.to_categorical(y_train)
+    y_train = np_utils.to_categorical(y_train)
 
     base_model = testing_utils.get_model_from_layers(
         [keras.layers.Dense(16,
@@ -141,7 +142,7 @@ class SequentialIntegrationTest(KerasIntegrationTest):
         test_samples=0,
         input_shape=(10,),
         num_classes=2)
-    y_train = keras.utils.to_categorical(y_train)
+    y_train = np_utils.to_categorical(y_train)
     model = keras.Sequential([
         keras.layers.Dense(16, activation='relu'),
         keras.layers.Dropout(0.1),
@@ -199,7 +200,7 @@ class TimeseriesClassificationIntegrationTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(4, 10),
         num_classes=2)
-    y_train = keras.utils.to_categorical(y_train)
+    y_train = np_utils.to_categorical(y_train)
 
     layers = [
         keras.layers.LSTM(5, return_sequences=True),
@@ -229,7 +230,7 @@ class TimeseriesClassificationIntegrationTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(4, 10),
         num_classes=2)
-    y_train = keras.utils.to_categorical(y_train)
+    y_train = np_utils.to_categorical(y_train)
 
     model = keras.models.Sequential()
     model.add(keras.layers.RNN(rnn_cell.LSTMCell(5), return_sequences=True,
@@ -264,7 +265,7 @@ class ImageClassificationIntegrationTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(10, 10, 3),
         num_classes=2)
-    y_train = keras.utils.to_categorical(y_train)
+    y_train = np_utils.to_categorical(y_train)
 
     layers = [
         keras.layers.Conv2D(4, 3, padding='same', activation='relu'),
@@ -308,7 +309,7 @@ class ActivationV2IntegrationTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(10,),
         num_classes=2)
-    y_train = keras.utils.to_categorical(y_train)
+    y_train = np_utils.to_categorical(y_train)
 
     model = keras.Sequential([
         keras.layers.Flatten(input_shape=x_train.shape[1:]),
diff --git a/tensorflow/python/keras/layers/gru_test.py b/tensorflow/python/keras/layers/gru_test.py
index cf32b8022b7..465ecc8f18f 100644
--- a/tensorflow/python/keras/layers/gru_test.py
+++ b/tensorflow/python/keras/layers/gru_test.py
@@ -26,6 +26,7 @@ from tensorflow.python.eager import context
 from tensorflow.python.framework import test_util as tf_test_util
 from tensorflow.python.keras import keras_parameterized
 from tensorflow.python.keras import testing_utils
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.platform import test
 
 
@@ -108,7 +109,7 @@ class GRULayerTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(timesteps, embedding_dim),
         num_classes=units)
-    y_train = keras.utils.to_categorical(y_train, units)
+    y_train = np_utils.to_categorical(y_train, units)
 
     inputs = keras.layers.Input(shape=[timesteps, embedding_dim])
     gru_layer = keras.layers.GRU(units,
diff --git a/tensorflow/python/keras/layers/gru_v2_test.py b/tensorflow/python/keras/layers/gru_v2_test.py
index 0a58879bdd1..013509102e3 100644
--- a/tensorflow/python/keras/layers/gru_v2_test.py
+++ b/tensorflow/python/keras/layers/gru_v2_test.py
@@ -38,6 +38,7 @@ from tensorflow.python.keras import keras_parameterized
 from tensorflow.python.keras import testing_utils
 from tensorflow.python.keras.layers import recurrent as rnn_v1
 from tensorflow.python.keras.layers import recurrent_v2 as rnn
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import control_flow_ops
 from tensorflow.python.ops import gen_math_ops
@@ -91,7 +92,7 @@ class GRUV2Test(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(timestep, input_shape),
         num_classes=output_shape)
-    y_train = keras.utils.to_categorical(y_train, output_shape)
+    y_train = np_utils.to_categorical(y_train, output_shape)
 
     layer = rnn.GRU(rnn_state_size)
 
@@ -150,7 +151,7 @@ class GRUV2Test(keras_parameterized.TestCase):
         input_shape=(timestep, input_shape),
         num_classes=rnn_state_size,
         random_seed=random_seed.DEFAULT_GRAPH_SEED)
-    y_train = keras.utils.to_categorical(y_train, rnn_state_size)
+    y_train = np_utils.to_categorical(y_train, rnn_state_size)
     # For the last batch item of the test data, we filter out the last
     # timestep to simulate the variable length sequence and masking test.
     x_train[-2:, -1, :] = 0.0
@@ -619,7 +620,7 @@ class GRUGraphRewriteTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(self.timestep, self.input_shape),
         num_classes=self.output_shape)
-    y_train = keras.utils.to_categorical(y_train, self.output_shape)
+    y_train = np_utils.to_categorical(y_train, self.output_shape)
 
     model.compile(
         optimizer='sgd',
@@ -679,7 +680,7 @@ class GRUGraphRewriteTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(self.timestep, self.input_shape),
         num_classes=self.output_shape)
-    y_train = keras.utils.to_categorical(y_train, self.output_shape)
+    y_train = np_utils.to_categorical(y_train, self.output_shape)
 
     model.compile(
         optimizer='sgd',
diff --git a/tensorflow/python/keras/layers/lstm_v2_test.py b/tensorflow/python/keras/layers/lstm_v2_test.py
index aed9f0c85fd..20f21c86fe0 100644
--- a/tensorflow/python/keras/layers/lstm_v2_test.py
+++ b/tensorflow/python/keras/layers/lstm_v2_test.py
@@ -39,6 +39,7 @@ from tensorflow.python.keras import keras_parameterized
 from tensorflow.python.keras import testing_utils
 from tensorflow.python.keras.layers import recurrent as rnn_v1
 from tensorflow.python.keras.layers import recurrent_v2 as rnn
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import control_flow_ops
 from tensorflow.python.ops import gen_math_ops
@@ -323,7 +324,7 @@ class LSTMV2Test(keras_parameterized.TestCase):
         input_shape=(timestep, input_shape),
         num_classes=rnn_state_size,
         random_seed=random_seed.DEFAULT_GRAPH_SEED)
-    y_train = keras.utils.to_categorical(y_train, rnn_state_size)
+    y_train = np_utils.to_categorical(y_train, rnn_state_size)
     # For the last batch item of the test data, we filter out the last
     # timestep to simulate the variable length sequence and masking test.
     x_train[-2:, -1, :] = 0.0
@@ -465,7 +466,7 @@ class LSTMV2Test(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(timestep, input_shape),
         num_classes=output_shape)
-    y_train = keras.utils.to_categorical(y_train, output_shape)
+    y_train = np_utils.to_categorical(y_train, output_shape)
 
     layer = rnn.LSTM(rnn_state_size)
 
@@ -799,7 +800,7 @@ class LSTMGraphRewriteTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(self.timestep, self.input_shape),
         num_classes=self.output_shape)
-    y_train = keras.utils.to_categorical(y_train, self.output_shape)
+    y_train = np_utils.to_categorical(y_train, self.output_shape)
 
     model.compile(
         optimizer='sgd',
@@ -860,7 +861,7 @@ class LSTMGraphRewriteTest(keras_parameterized.TestCase):
         test_samples=0,
         input_shape=(self.timestep, self.input_shape),
         num_classes=self.output_shape)
-    y_train = keras.utils.to_categorical(y_train, self.output_shape)
+    y_train = np_utils.to_categorical(y_train, self.output_shape)
 
     model.compile(
         optimizer='sgd',
@@ -1020,7 +1021,7 @@ class LSTMPerformanceTest(test.Benchmark):
         test_samples=0,
         input_shape=(test_config['timestep'], test_config['input_shape']),
         num_classes=test_config['output_shape'])
-    y_train = keras.utils.to_categorical(y_train, test_config['output_shape'])
+    y_train = np_utils.to_categorical(y_train, test_config['output_shape'])
 
     cudnn_sec_per_epoch = self._time_performance_run_cudnn_lstm(
         test_config, x_train, y_train)
diff --git a/tensorflow/python/keras/layers/recurrent_test.py b/tensorflow/python/keras/layers/recurrent_test.py
index 57d4a99938e..af222f366a4 100644
--- a/tensorflow/python/keras/layers/recurrent_test.py
+++ b/tensorflow/python/keras/layers/recurrent_test.py
@@ -38,6 +38,7 @@ from tensorflow.python.keras import testing_utils
 from tensorflow.python.keras.engine import base_layer_utils
 from tensorflow.python.keras.layers import recurrent as rnn_v1
 from tensorflow.python.keras.layers import recurrent_v2 as rnn_v2
+from tensorflow.python.keras.utils import generic_utils
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import init_ops
 from tensorflow.python.ops import math_ops
@@ -196,7 +197,7 @@ class RNNTest(keras_parameterized.TestCase):
     y_np = model.predict(x_np)
     weights = model.get_weights()
     config = layer.get_config()
-    with keras.utils.CustomObjectScope({'MinimalRNNCell': MinimalRNNCell}):
+    with generic_utils.CustomObjectScope({'MinimalRNNCell': MinimalRNNCell}):
       layer = keras.layers.RNN.from_config(config)
     y = layer(x)
     model = keras.models.Model(x, y)
@@ -223,7 +224,7 @@ class RNNTest(keras_parameterized.TestCase):
     y_np = model.predict(x_np)
     weights = model.get_weights()
     config = layer.get_config()
-    with keras.utils.CustomObjectScope({'MinimalRNNCell': MinimalRNNCell}):
+    with generic_utils.CustomObjectScope({'MinimalRNNCell': MinimalRNNCell}):
       layer = keras.layers.RNN.from_config(config)
     y = layer(x)
     model = keras.models.Model(x, y)
@@ -417,7 +418,7 @@ class RNNTest(keras_parameterized.TestCase):
     weights = model.get_weights()
     config = layer.get_config()
     custom_objects = {'RNNCellWithConstants': RNNCellWithConstants}
-    with keras.utils.CustomObjectScope(custom_objects):
+    with generic_utils.CustomObjectScope(custom_objects):
       layer = keras.layers.RNN.from_config(config.copy())
     y = layer(x, constants=c)
     model = keras.models.Model([x, c], y)
@@ -426,7 +427,7 @@ class RNNTest(keras_parameterized.TestCase):
     self.assertAllClose(y_np, y_np_2, atol=1e-4)
 
     # test flat list inputs.
-    with keras.utils.CustomObjectScope(custom_objects):
+    with generic_utils.CustomObjectScope(custom_objects):
       layer = keras.layers.RNN.from_config(config.copy())
     y = layer([x, c])
     model = keras.models.Model([x, c], y)
@@ -474,7 +475,7 @@ class RNNTest(keras_parameterized.TestCase):
     y_np = model.predict([x_np, c_np])
     weights = model.get_weights()
     config = layer.get_config()
-    with keras.utils.CustomObjectScope(custom_objects):
+    with generic_utils.CustomObjectScope(custom_objects):
       layer = keras.layers.recurrent.RNN.from_config(config.copy())
     y = layer(x, constants=c)
     model = keras.models.Model([x, c], y)
@@ -539,7 +540,7 @@ class RNNTest(keras_parameterized.TestCase):
     weights = model.get_weights()
     config = layer.get_config()
     custom_objects = {'RNNCellWithConstants': RNNCellWithConstants}
-    with keras.utils.CustomObjectScope(custom_objects):
+    with generic_utils.CustomObjectScope(custom_objects):
       layer = keras.layers.RNN.from_config(config.copy())
     y = layer(x, initial_state=s, constants=c)
     model = keras.models.Model([x, s, c], y)
@@ -553,7 +554,7 @@ class RNNTest(keras_parameterized.TestCase):
       self.assertAllClose(y_np, y_np_2_different_s, atol=1e-4)
 
     # test flat list inputs
-    with keras.utils.CustomObjectScope(custom_objects):
+    with generic_utils.CustomObjectScope(custom_objects):
       layer = keras.layers.RNN.from_config(config.copy())
     y = layer([x, s, c])
     model = keras.models.Model([x, s, c], y)
diff --git a/tensorflow/python/keras/layers/wrappers_test.py b/tensorflow/python/keras/layers/wrappers_test.py
index d8f4071c488..8c81a105088 100644
--- a/tensorflow/python/keras/layers/wrappers_test.py
+++ b/tensorflow/python/keras/layers/wrappers_test.py
@@ -32,6 +32,7 @@ from tensorflow.python.keras import keras_parameterized
 from tensorflow.python.keras import testing_utils
 from tensorflow.python.keras.engine import base_layer_utils
 from tensorflow.python.keras.layers.rnn_cell_wrapper_v2 import ResidualWrapper
+from tensorflow.python.keras.utils import generic_utils
 from tensorflow.python.ops.array_ops import concat
 from tensorflow.python.platform import test
 from tensorflow.python.training.tracking import util as trackable_util
@@ -685,7 +686,7 @@ class BidirectionalTest(test.TestCase, parameterized.TestCase):
       c = keras.Input((3,))
       cell = _RNNCellWithConstants(32, 3)
       custom_objects = {'_RNNCellWithConstants': _RNNCellWithConstants}
-      with keras.utils.CustomObjectScope(custom_objects):
+      with generic_utils.CustomObjectScope(custom_objects):
         layer = keras.layers.Bidirectional(keras.layers.RNN(cell))
       y = layer(x, constants=c)
       model = keras.Model([x, c], y)
@@ -702,7 +703,7 @@ class BidirectionalTest(test.TestCase, parameterized.TestCase):
       weights = model.get_weights()
       config = layer.get_config()
 
-      with keras.utils.CustomObjectScope(custom_objects):
+      with generic_utils.CustomObjectScope(custom_objects):
         layer = keras.layers.Bidirectional.from_config(copy.deepcopy(config))
       y = layer(x, constants=c)
       model = keras.Model([x, c], y)
@@ -711,7 +712,7 @@ class BidirectionalTest(test.TestCase, parameterized.TestCase):
       self.assertAllClose(y_np, y_np_2, atol=1e-4)
 
       # Test flat list inputs
-      with keras.utils.CustomObjectScope(custom_objects):
+      with generic_utils.CustomObjectScope(custom_objects):
         layer = keras.layers.Bidirectional.from_config(copy.deepcopy(config))
       y = layer([x, c])
       model = keras.Model([x, c], y)
@@ -728,7 +729,7 @@ class BidirectionalTest(test.TestCase, parameterized.TestCase):
       s_bac = keras.Input((32,))
       cell = _RNNCellWithConstants(32, 3)
       custom_objects = {'_RNNCellWithConstants': _RNNCellWithConstants}
-      with keras.utils.CustomObjectScope(custom_objects):
+      with generic_utils.CustomObjectScope(custom_objects):
         layer = keras.layers.Bidirectional(keras.layers.RNN(cell))
       y = layer(x, initial_state=[s_for, s_bac], constants=c)
       model = keras.Model([x, s_for, s_bac, c], y)
@@ -750,7 +751,7 @@ class BidirectionalTest(test.TestCase, parameterized.TestCase):
       weights = model.get_weights()
       config = layer.get_config()
 
-      with keras.utils.CustomObjectScope(custom_objects):
+      with generic_utils.CustomObjectScope(custom_objects):
         layer = keras.layers.Bidirectional.from_config(copy.deepcopy(config))
       y = layer(x, initial_state=[s_for, s_bac], constants=c)
       model = keras.Model([x, s_for, s_bac, c], y)
@@ -764,7 +765,7 @@ class BidirectionalTest(test.TestCase, parameterized.TestCase):
       assert np.mean(y_np - y_np_2_different_s) != 0
 
       # Test flat list inputs
-      with keras.utils.CustomObjectScope(custom_objects):
+      with generic_utils.CustomObjectScope(custom_objects):
         layer = keras.layers.Bidirectional.from_config(copy.deepcopy(config))
       y = layer([x, s_for, s_bac, c])
       model = keras.Model([x, s_for, s_bac, c], y)
diff --git a/tensorflow/python/keras/losses_test.py b/tensorflow/python/keras/losses_test.py
index 0cb67f609d0..8cd7b320956 100644
--- a/tensorflow/python/keras/losses_test.py
+++ b/tensorflow/python/keras/losses_test.py
@@ -28,6 +28,7 @@ from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
 from tensorflow.python.framework import test_util
+from tensorflow.python.keras.utils import generic_utils
 from tensorflow.python.keras.utils import losses_utils
 from tensorflow.python.platform import test
 
@@ -200,10 +201,10 @@ class KerasLossesTest(test.TestCase):
 
   def test_serializing_loss_class(self):
     orig_loss_class = _MSEMAELoss(0.3)
-    with keras.utils.custom_object_scope({'_MSEMAELoss': _MSEMAELoss}):
+    with generic_utils.custom_object_scope({'_MSEMAELoss': _MSEMAELoss}):
       serialized = keras.losses.serialize(orig_loss_class)
 
-    with keras.utils.custom_object_scope({'_MSEMAELoss': _MSEMAELoss}):
+    with generic_utils.custom_object_scope({'_MSEMAELoss': _MSEMAELoss}):
       deserialized = keras.losses.deserialize(serialized)
     assert isinstance(deserialized, _MSEMAELoss)
     assert deserialized.mse_fraction == 0.3
@@ -214,7 +215,7 @@ class KerasLossesTest(test.TestCase):
     model_filename = os.path.join(tmpdir, 'custom_loss.h5')
 
     with self.cached_session():
-      with keras.utils.custom_object_scope({'_MSEMAELoss': _MSEMAELoss}):
+      with generic_utils.custom_object_scope({'_MSEMAELoss': _MSEMAELoss}):
         loss = _MSEMAELoss(0.3)
         inputs = keras.layers.Input((2,))
         outputs = keras.layers.Dense(1, name='model_output')(inputs)
@@ -227,7 +228,7 @@ class KerasLossesTest(test.TestCase):
 
         model.save(model_filename)
 
-      with keras.utils.custom_object_scope({'_MSEMAELoss': _MSEMAELoss}):
+      with generic_utils.custom_object_scope({'_MSEMAELoss': _MSEMAELoss}):
         loaded_model = keras.models.load_model(model_filename)
         loaded_model.predict(np.random.rand(128, 2))
 
diff --git a/tensorflow/python/keras/optimizer_v2/optimizer_v2_test.py b/tensorflow/python/keras/optimizer_v2/optimizer_v2_test.py
index 3868a6fb16c..1caa4dc85de 100644
--- a/tensorflow/python/keras/optimizer_v2/optimizer_v2_test.py
+++ b/tensorflow/python/keras/optimizer_v2/optimizer_v2_test.py
@@ -50,6 +50,7 @@ from tensorflow.python.keras.optimizer_v2 import learning_rate_schedule
 from tensorflow.python.keras.optimizer_v2 import nadam
 from tensorflow.python.keras.optimizer_v2 import optimizer_v2
 from tensorflow.python.keras.optimizer_v2 import rmsprop
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import clip_ops
 from tensorflow.python.ops import resource_variable_ops
@@ -556,7 +557,7 @@ class OptimizerTest(test.TestCase):
         test_samples=10,
         input_shape=(input_dim,),
         num_classes=num_classes)
-    y = keras.utils.to_categorical(y)
+    y = np_utils.to_categorical(y)
 
     num_hidden = 1
     model = testing_utils.get_small_sequential_mlp(
@@ -635,7 +636,7 @@ class OptimizersCompatibilityTest(keras_parameterized.TestCase):
           test_samples=10,
           input_shape=(input_dim,),
           num_classes=num_classes)
-      y = keras.utils.to_categorical(y)
+      y = np_utils.to_categorical(y)
 
       num_hidden = 5
       model_v1 = testing_utils.get_small_sequential_mlp(
@@ -723,7 +724,7 @@ class OptimizersCompatibilityTest(keras_parameterized.TestCase):
           test_samples=10,
           input_shape=(input_dim,),
           num_classes=num_classes)
-      y = keras.utils.to_categorical(y)
+      y = np_utils.to_categorical(y)
 
       num_hidden = 5
       model_k_v1 = testing_utils.get_small_sequential_mlp(
@@ -784,7 +785,7 @@ class OptimizersCompatibilityTest(keras_parameterized.TestCase):
           test_samples=10,
           input_shape=(input_dim,),
           num_classes=num_classes)
-      y = keras.utils.to_categorical(y)
+      y = np_utils.to_categorical(y)
 
       num_hidden = 5
       model_k_v1 = testing_utils.get_small_sequential_mlp(
diff --git a/tensorflow/python/keras/optimizers_test.py b/tensorflow/python/keras/optimizers_test.py
index 8885cead8b5..67b37965457 100644
--- a/tensorflow/python/keras/optimizers_test.py
+++ b/tensorflow/python/keras/optimizers_test.py
@@ -28,6 +28,7 @@ from tensorflow.python.eager import context
 from tensorflow.python.framework import ops
 from tensorflow.python.keras import keras_parameterized
 from tensorflow.python.keras import testing_utils
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.platform import test
 from tensorflow.python.training.adam import AdamOptimizer
 
@@ -54,7 +55,7 @@ class KerasOptimizersTest(keras_parameterized.TestCase):
     np.random.seed(1337)
     (x_train, y_train), _ = testing_utils.get_test_data(
         train_samples=1000, test_samples=200, input_shape=(10,), num_classes=2)
-    y_train = keras.utils.to_categorical(y_train)
+    y_train = np_utils.to_categorical(y_train)
     model = _get_model(x_train.shape[1], 20, y_train.shape[1])
     model.compile(
         loss='categorical_crossentropy',
diff --git a/tensorflow/python/keras/preprocessing/__init__.py b/tensorflow/python/keras/preprocessing/__init__.py
index 0860eed3cf9..3aba38b2341 100644
--- a/tensorflow/python/keras/preprocessing/__init__.py
+++ b/tensorflow/python/keras/preprocessing/__init__.py
@@ -18,15 +18,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import keras_preprocessing
-
-from tensorflow.python.keras import backend
-from tensorflow.python.keras import utils
-
-# This exists for compatibility with prior version of keras_preprocessing.
-# TODO(fchollet): remove in the future.
-keras_preprocessing.set_keras_submodules(backend=backend, utils=utils)
-
 from tensorflow.python.keras.preprocessing import image
 from tensorflow.python.keras.preprocessing import sequence
 from tensorflow.python.keras.preprocessing import text
diff --git a/tensorflow/python/keras/preprocessing/image.py b/tensorflow/python/keras/preprocessing/image.py
index f2fefffb0a8..25965cdf239 100644
--- a/tensorflow/python/keras/preprocessing/image.py
+++ b/tensorflow/python/keras/preprocessing/image.py
@@ -28,7 +28,7 @@ except ImportError:
   pass
 
 from tensorflow.python.keras import backend
-from tensorflow.python.keras import utils
+from tensorflow.python.keras.utils import data_utils
 from tensorflow.python.util import tf_inspect
 from tensorflow.python.util.tf_export import keras_export
 
@@ -132,7 +132,7 @@ def save_img(path,
 
 
 @keras_export('keras.preprocessing.image.Iterator')
-class Iterator(image.Iterator, utils.Sequence):
+class Iterator(image.Iterator, data_utils.Sequence):
   pass
 
 
diff --git a/tensorflow/python/keras/preprocessing/sequence.py b/tensorflow/python/keras/preprocessing/sequence.py
index 1d73a1e4da5..41a3ab6c1be 100644
--- a/tensorflow/python/keras/preprocessing/sequence.py
+++ b/tensorflow/python/keras/preprocessing/sequence.py
@@ -21,7 +21,7 @@ from __future__ import print_function
 
 from keras_preprocessing import sequence
 
-from tensorflow.python.keras import utils
+from tensorflow.python.keras.utils import data_utils
 from tensorflow.python.util.tf_export import keras_export
 
 pad_sequences = sequence.pad_sequences
@@ -32,7 +32,7 @@ _remove_long_seq = sequence._remove_long_seq  # pylint: disable=protected-access
 
 
 @keras_export('keras.preprocessing.sequence.TimeseriesGenerator')
-class TimeseriesGenerator(sequence.TimeseriesGenerator, utils.Sequence):
+class TimeseriesGenerator(sequence.TimeseriesGenerator, data_utils.Sequence):
   """Utility class for generating batches of temporal data.
   This class takes in a sequence of data-points gathered at
   equal intervals, along with time series parameters such as
diff --git a/tensorflow/python/keras/regularizers_test.py b/tensorflow/python/keras/regularizers_test.py
index cabefa2694b..c8180ca462b 100644
--- a/tensorflow/python/keras/regularizers_test.py
+++ b/tensorflow/python/keras/regularizers_test.py
@@ -26,6 +26,7 @@ from tensorflow.python.eager import context
 from tensorflow.python.keras import keras_parameterized
 from tensorflow.python.keras import regularizers
 from tensorflow.python.keras import testing_utils
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.ops import math_ops
 from tensorflow.python.platform import test
 
@@ -51,8 +52,8 @@ class KerasRegularizersTest(keras_parameterized.TestCase,
         test_samples=10,
         input_shape=(DATA_DIM,),
         num_classes=NUM_CLASSES)
-    y_train = keras.utils.to_categorical(y_train, NUM_CLASSES)
-    y_test = keras.utils.to_categorical(y_test, NUM_CLASSES)
+    y_train = np_utils.to_categorical(y_train, NUM_CLASSES)
+    y_test = np_utils.to_categorical(y_test, NUM_CLASSES)
     return (x_train, y_train), (x_test, y_test)
 
   def create_multi_input_model_from(self, layer1, layer2):
diff --git a/tensorflow/python/keras/utils/__init__.py b/tensorflow/python/keras/utils/__init__.py
index 3995391a6ef..e69de29bb2d 100644
--- a/tensorflow/python/keras/utils/__init__.py
+++ b/tensorflow/python/keras/utils/__init__.py
@@ -1,46 +0,0 @@
-# Copyright 2016 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.
-# ==============================================================================
-"""Keras utilities."""
-
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-
-from tensorflow.python.keras.utils.data_utils import GeneratorEnqueuer
-from tensorflow.python.keras.utils.data_utils import get_file
-from tensorflow.python.keras.utils.data_utils import OrderedEnqueuer
-from tensorflow.python.keras.utils.data_utils import Sequence
-from tensorflow.python.keras.utils.data_utils import SequenceEnqueuer
-from tensorflow.python.keras.utils.generic_utils import class_and_config_for_serialized_keras_object
-from tensorflow.python.keras.utils.generic_utils import custom_object_scope
-from tensorflow.python.keras.utils.generic_utils import CustomObjectScope
-from tensorflow.python.keras.utils.generic_utils import deserialize_keras_object
-from tensorflow.python.keras.utils.generic_utils import get_custom_objects
-from tensorflow.python.keras.utils.generic_utils import Progbar
-from tensorflow.python.keras.utils.generic_utils import serialize_keras_class_and_config
-from tensorflow.python.keras.utils.generic_utils import serialize_keras_object
-from tensorflow.python.keras.utils.io_utils import HDF5Matrix
-from tensorflow.python.keras.utils.layer_utils import convert_all_kernels_in_model
-from tensorflow.python.keras.utils.layer_utils import get_source_inputs
-from tensorflow.python.keras.utils.layer_utils import print_summary
-from tensorflow.python.keras.utils.multi_gpu_utils import multi_gpu_model
-from tensorflow.python.keras.utils.np_utils import normalize
-from tensorflow.python.keras.utils.np_utils import to_categorical
-from tensorflow.python.keras.utils.vis_utils import model_to_dot
-from tensorflow.python.keras.utils.vis_utils import plot_model
-
-del absolute_import
-del division
-del print_function
diff --git a/tensorflow/python/keras/utils/all_utils.py b/tensorflow/python/keras/utils/all_utils.py
new file mode 100644
index 00000000000..deb73b08337
--- /dev/null
+++ b/tensorflow/python/keras/utils/all_utils.py
@@ -0,0 +1,48 @@
+# Copyright 2019 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.
+# ==============================================================================
+"""Public Keras utilities.
+
+This module is used as a shortcut to access all the symbols. Those symbols was
+exposed under __init__, and was causing some hourglass import issue.
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+# pylint: disable=unused-import
+from tensorflow.python.keras.utils.data_utils import GeneratorEnqueuer
+from tensorflow.python.keras.utils.data_utils import get_file
+from tensorflow.python.keras.utils.data_utils import OrderedEnqueuer
+from tensorflow.python.keras.utils.data_utils import Sequence
+from tensorflow.python.keras.utils.data_utils import SequenceEnqueuer
+from tensorflow.python.keras.utils.generic_utils import custom_object_scope
+from tensorflow.python.keras.utils.generic_utils import CustomObjectScope
+from tensorflow.python.keras.utils.generic_utils import deserialize_keras_object
+from tensorflow.python.keras.utils.generic_utils import get_custom_objects
+from tensorflow.python.keras.utils.generic_utils import Progbar
+from tensorflow.python.keras.utils.generic_utils import serialize_keras_object
+from tensorflow.python.keras.utils.io_utils import HDF5Matrix
+from tensorflow.python.keras.utils.layer_utils import convert_all_kernels_in_model
+from tensorflow.python.keras.utils.layer_utils import get_source_inputs
+from tensorflow.python.keras.utils.multi_gpu_utils import multi_gpu_model
+from tensorflow.python.keras.utils.np_utils import normalize
+from tensorflow.python.keras.utils.np_utils import to_categorical
+from tensorflow.python.keras.utils.vis_utils import model_to_dot
+from tensorflow.python.keras.utils.vis_utils import plot_model
+
+del absolute_import
+del division
+del print_function
diff --git a/tensorflow/python/keras/utils/multi_gpu_utils_test.py b/tensorflow/python/keras/utils/multi_gpu_utils_test.py
index 38f53c6ae83..5b3f9bcd8d7 100644
--- a/tensorflow/python/keras/utils/multi_gpu_utils_test.py
+++ b/tensorflow/python/keras/utils/multi_gpu_utils_test.py
@@ -21,6 +21,7 @@ import numpy as np
 
 from tensorflow.python import data
 from tensorflow.python import keras
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.platform import test
 
 
@@ -149,7 +150,7 @@ class TestMultiGPUModel(test.TestCase):
       x_train = np.random.randint(0, 255, input_shape)
       y_train = np.random.randint(0, num_classes, (input_shape[0],))
 
-      y_train = keras.utils.to_categorical(y_train, num_classes)
+      y_train = np_utils.to_categorical(y_train, num_classes)
 
       x_train = x_train.astype('float32')
       y_train = y_train.astype('float32')
diff --git a/tensorflow/python/keras/utils/np_utils_test.py b/tensorflow/python/keras/utils/np_utils_test.py
index 8e0e127cc3b..d79e6afd81f 100644
--- a/tensorflow/python/keras/utils/np_utils_test.py
+++ b/tensorflow/python/keras/utils/np_utils_test.py
@@ -20,7 +20,7 @@ from __future__ import print_function
 
 import numpy as np
 
-from tensorflow.python import keras
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.platform import test
 
 
@@ -34,7 +34,7 @@ class TestNPUtils(test.TestCase):
                        (3, 2, num_classes)]
     labels = [np.random.randint(0, num_classes, shape) for shape in shapes]
     one_hots = [
-        keras.utils.to_categorical(label, num_classes) for label in labels]
+        np_utils.to_categorical(label, num_classes) for label in labels]
     for label, one_hot, expected_shape in zip(labels,
                                               one_hots,
                                               expected_shapes):
diff --git a/tensorflow/python/kernel_tests/rnn_test.py b/tensorflow/python/kernel_tests/rnn_test.py
index bdb01700582..7b35d7338b1 100644
--- a/tensorflow/python/kernel_tests/rnn_test.py
+++ b/tensorflow/python/kernel_tests/rnn_test.py
@@ -36,6 +36,7 @@ from tensorflow.python.framework import tensor_shape
 from tensorflow.python.framework import test_util
 from tensorflow.python.keras import testing_utils
 from tensorflow.python.keras.engine import network as keras_network
+from tensorflow.python.keras.utils import np_utils
 from tensorflow.python.layers import base as base_layers
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import control_flow_ops
@@ -372,7 +373,7 @@ class RNNTest(test.TestCase):
           test_samples=0,
           input_shape=(timestep, input_shape),
           num_classes=output_shape)
-      y_train = keras.utils.to_categorical(y_train)
+      y_train = np_utils.to_categorical(y_train)
       cell = keras.layers.SimpleRNNCell(output_shape)
 
       inputs = array_ops.placeholder(
@@ -406,7 +407,7 @@ class RNNTest(test.TestCase):
           test_samples=0,
           input_shape=(timestep, input_shape),
           num_classes=output_shape)
-      y_train = keras.utils.to_categorical(y_train)
+      y_train = np_utils.to_categorical(y_train)
       cell = keras.layers.GRUCell(output_shape)
 
       inputs = array_ops.placeholder(
@@ -440,7 +441,7 @@ class RNNTest(test.TestCase):
           test_samples=0,
           input_shape=(timestep, input_shape),
           num_classes=output_shape)
-      y_train = keras.utils.to_categorical(y_train)
+      y_train = np_utils.to_categorical(y_train)
       cell = keras.layers.LSTMCell(output_shape)
 
       inputs = array_ops.placeholder(
@@ -478,7 +479,7 @@ class RNNTest(test.TestCase):
           test_samples=0,
           input_shape=(timestep, input_shape),
           num_classes=output_shape)
-      y_train = keras.utils.to_categorical(y_train)
+      y_train = np_utils.to_categorical(y_train)
       cell = keras.layers.StackedRNNCells(
           [keras.layers.LSTMCell(2 * output_shape),
            keras.layers.LSTMCell(output_shape)])
@@ -523,7 +524,7 @@ class RNNTest(test.TestCase):
           input_shape=(timestep, input_shape),
           num_classes=output_shape)
       x_train = np.transpose(x_train, (1, 0, 2))
-      y_train = keras.utils.to_categorical(y_train)
+      y_train = np_utils.to_categorical(y_train)
       cell = keras.layers.SimpleRNNCell(output_shape)
 
       inputs = [array_ops.placeholder(