Merged commit includes the following changes: (tensorflow#8077)

Internal cleanup (py2->py3) plus the following changes: 285513318 by Sergio Guadarrama: Adds a script for post-training quantization 284222305 by Sergio Guadarrama: Modified squeeze-excite operation to accommodate tensors of undefined (Nonetype) H/W. 282028343 by Sergio Guadarrama: Add MobilenetV3 and MobilenetEdgeTPU to the slim/nets_factory. PiperOrigin-RevId: 289455329 Co-authored-by: Sergio Guadarrama <[email protected]>
aadaa88 · Jan 22, 2020 · d4eedbb · d4eedbb
1 parent 0e0a94a
commit d4eedbb
Show file tree

Hide file tree

Showing 58 changed files with 1,436 additions and 1,071 deletions.
diff --git a/research/slim/BUILD b/research/slim/BUILD
@@ -1,6 +1,7 @@
 # Description:
 #   Contains files for loading, training and evaluating TF-Slim-based models.
 # load("//devtools/python/blaze:python3.bzl", "py2and3_test")
+load("//devtools/python/blaze:pytype.bzl", "pytype_strict_binary")
 
 package(
     default_visibility = ["//visibility:public"],
@@ -475,11 +476,10 @@ py_test(
     ],
 )
 
-py_test(
+py_test(  # py2and3_test
     name = "inception_v2_test",
     size = "large",
     srcs = ["nets/inception_v2_test.py"],
-    python_version = "PY2",
     shard_count = 3,
     srcs_version = "PY2AND3",
     deps = [
@@ -590,14 +590,14 @@ py_library(
     ],
 )
 
-py_test(
+py_test(  # py2and3_test
     name = "mobilenet_v2_test",
     srcs = ["nets/mobilenet/mobilenet_v2_test.py"],
-    python_version = "PY2",
     srcs_version = "PY2AND3",
     deps = [
         ":mobilenet",
         ":mobilenet_common",
+        "//third_party/py/six",
         # "//tensorflow",
         # "//tensorflow/contrib/slim",
     ],
@@ -755,11 +755,10 @@ py_library(
     ],
 )
 
-py_test(
+py_test(  # py2and3_test
     name = "overfeat_test",
     size = "medium",
     srcs = ["nets/overfeat_test.py"],
-    python_version = "PY2",
     srcs_version = "PY2AND3",
     deps = [
         ":overfeat",
@@ -890,11 +889,10 @@ py_library(
     ],
 )
 
-py_test(
+py_test(  # py2and3_test
     name = "vgg_test",
     size = "medium",
     srcs = ["nets/vgg_test.py"],
-    python_version = "PY2",
     srcs_version = "PY2AND3",
     deps = [
         ":vgg",
@@ -912,11 +910,10 @@ py_library(
     ],
 )
 
-py_test(
+py_test(  # py2and3_test
     name = "nets_factory_test",
     size = "large",
     srcs = ["nets/nets_factory_test.py"],
-    python_version = "PY2",
     shard_count = 3,
     srcs_version = "PY2AND3",
     deps = [
@@ -925,9 +922,24 @@ py_test(
     ],
 )
 
+pytype_strict_binary(
+    name = "post_training_quantization",
+    srcs = ["nets/post_training_quantization.py"],
+    python_version = "PY3",
+    deps = [
+        ":nets_factory",
+        ":preprocessing_factory",
+        "//third_party/py/absl:app",
+        "//third_party/py/absl/flags",
+        # "//tensorflow",
+        # "//tensorflow_datasets",
+    ],
+)
+
 py_library(
     name = "train_image_classifier_lib",
     srcs = ["train_image_classifier.py"],
+    srcs_version = "PY2AND3",
     deps = [
         ":dataset_factory",
         ":model_deploy",

diff --git a/research/slim/datasets/download_and_convert_visualwakewords_lib.py b/research/slim/datasets/download_and_convert_visualwakewords_lib.py
@@ -201,7 +201,7 @@ def create_tf_record_for_visualwakewords_dataset(annotations_file, image_dir,
     groundtruth_data = json.load(fid)
     images = groundtruth_data['images']
     annotations_index = groundtruth_data['annotations']
-    annotations_index = {int(k): v for k, v in annotations_index.items()}
+    annotations_index = {int(k): v for k, v in annotations_index.iteritems()}
     # convert 'unicode' key to 'int' key after we parse the json file
 
     for idx, image in enumerate(images):

diff --git a/research/slim/nets/alexnet.py b/research/slim/nets/alexnet.py
@@ -40,13 +40,16 @@
 from tensorflow.contrib import slim as contrib_slim
 
 slim = contrib_slim
-trunc_normal = lambda stddev: tf.truncated_normal_initializer(0.0, stddev)
+
+# pylint: disable=g-long-lambda
+trunc_normal = lambda stddev: tf.compat.v1.truncated_normal_initializer(
+    0.0, stddev)
 
 
 def alexnet_v2_arg_scope(weight_decay=0.0005):
   with slim.arg_scope([slim.conv2d, slim.fully_connected],
                       activation_fn=tf.nn.relu,
-                      biases_initializer=tf.constant_initializer(0.1),
+                      biases_initializer=tf.compat.v1.constant_initializer(0.1),
                       weights_regularizer=slim.l2_regularizer(weight_decay)):
     with slim.arg_scope([slim.conv2d], padding='SAME'):
       with slim.arg_scope([slim.max_pool2d], padding='VALID') as arg_sc:
@@ -94,7 +97,7 @@ def alexnet_v2(inputs,
       or None).
     end_points: a dict of tensors with intermediate activations.
   """
-  with tf.variable_scope(scope, 'alexnet_v2', [inputs]) as sc:
+  with tf.compat.v1.variable_scope(scope, 'alexnet_v2', [inputs]) as sc:
     end_points_collection = sc.original_name_scope + '_end_points'
     # Collect outputs for conv2d, fully_connected and max_pool2d.
     with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.max_pool2d],
@@ -110,9 +113,10 @@ def alexnet_v2(inputs,
       net = slim.max_pool2d(net, [3, 3], 2, scope='pool5')
 
       # Use conv2d instead of fully_connected layers.
-      with slim.arg_scope([slim.conv2d],
-                          weights_initializer=trunc_normal(0.005),
-                          biases_initializer=tf.constant_initializer(0.1)):
+      with slim.arg_scope(
+          [slim.conv2d],
+          weights_initializer=trunc_normal(0.005),
+          biases_initializer=tf.compat.v1.constant_initializer(0.1)):
         net = slim.conv2d(net, 4096, [5, 5], padding='VALID',
                           scope='fc6')
         net = slim.dropout(net, dropout_keep_prob, is_training=is_training,
@@ -122,16 +126,19 @@ def alexnet_v2(inputs,
         end_points = slim.utils.convert_collection_to_dict(
             end_points_collection)
         if global_pool:
-          net = tf.reduce_mean(net, [1, 2], keep_dims=True, name='global_pool')
+          net = tf.reduce_mean(
+              input_tensor=net, axis=[1, 2], keepdims=True, name='global_pool')
           end_points['global_pool'] = net
         if num_classes:
           net = slim.dropout(net, dropout_keep_prob, is_training=is_training,
                              scope='dropout7')
-          net = slim.conv2d(net, num_classes, [1, 1],
-                            activation_fn=None,
-                            normalizer_fn=None,
-                            biases_initializer=tf.zeros_initializer(),
-                            scope='fc8')
+          net = slim.conv2d(
+              net,
+              num_classes, [1, 1],
+              activation_fn=None,
+              normalizer_fn=None,
+              biases_initializer=tf.compat.v1.zeros_initializer(),
+              scope='fc8')
           if spatial_squeeze:
             net = tf.squeeze(net, [1, 2], name='fc8/squeezed')
           end_points[sc.name + '/fc8'] = net

diff --git a/research/slim/nets/alexnet_test.py b/research/slim/nets/alexnet_test.py
@@ -32,7 +32,7 @@ def testBuild(self):
     height, width = 224, 224
     num_classes = 1000
     with self.test_session():
-      inputs = tf.random_uniform((batch_size, height, width, 3))
+      inputs = tf.random.uniform((batch_size, height, width, 3))
       logits, _ = alexnet.alexnet_v2(inputs, num_classes)
       self.assertEquals(logits.op.name, 'alexnet_v2/fc8/squeezed')
       self.assertListEqual(logits.get_shape().as_list(),
@@ -43,7 +43,7 @@ def testFullyConvolutional(self):
     height, width = 300, 400
     num_classes = 1000
     with self.test_session():
-      inputs = tf.random_uniform((batch_size, height, width, 3))
+      inputs = tf.random.uniform((batch_size, height, width, 3))
       logits, _ = alexnet.alexnet_v2(inputs, num_classes, spatial_squeeze=False)
       self.assertEquals(logits.op.name, 'alexnet_v2/fc8/BiasAdd')
       self.assertListEqual(logits.get_shape().as_list(),
@@ -54,7 +54,7 @@ def testGlobalPool(self):
     height, width = 256, 256
     num_classes = 1000
     with self.test_session():
-      inputs = tf.random_uniform((batch_size, height, width, 3))
+      inputs = tf.random.uniform((batch_size, height, width, 3))
       logits, _ = alexnet.alexnet_v2(inputs, num_classes, spatial_squeeze=False,
                                      global_pool=True)
       self.assertEquals(logits.op.name, 'alexnet_v2/fc8/BiasAdd')
@@ -66,7 +66,7 @@ def testEndPoints(self):
     height, width = 224, 224
     num_classes = 1000
     with self.test_session():
-      inputs = tf.random_uniform((batch_size, height, width, 3))
+      inputs = tf.random.uniform((batch_size, height, width, 3))
       _, end_points = alexnet.alexnet_v2(inputs, num_classes)
       expected_names = ['alexnet_v2/conv1',
                         'alexnet_v2/pool1',
@@ -87,7 +87,7 @@ def testNoClasses(self):
     height, width = 224, 224
     num_classes = None
     with self.test_session():
-      inputs = tf.random_uniform((batch_size, height, width, 3))
+      inputs = tf.random.uniform((batch_size, height, width, 3))
       net, end_points = alexnet.alexnet_v2(inputs, num_classes)
       expected_names = ['alexnet_v2/conv1',
                         'alexnet_v2/pool1',
@@ -110,7 +110,7 @@ def testModelVariables(self):
     height, width = 224, 224
     num_classes = 1000
     with self.test_session():
-      inputs = tf.random_uniform((batch_size, height, width, 3))
+      inputs = tf.random.uniform((batch_size, height, width, 3))
       alexnet.alexnet_v2(inputs, num_classes)
       expected_names = ['alexnet_v2/conv1/weights',
                         'alexnet_v2/conv1/biases',
@@ -137,11 +137,11 @@ def testEvaluation(self):
     height, width = 224, 224
     num_classes = 1000
     with self.test_session():
-      eval_inputs = tf.random_uniform((batch_size, height, width, 3))
+      eval_inputs = tf.random.uniform((batch_size, height, width, 3))
       logits, _ = alexnet.alexnet_v2(eval_inputs, is_training=False)
       self.assertListEqual(logits.get_shape().as_list(),
                            [batch_size, num_classes])
-      predictions = tf.argmax(logits, 1)
+      predictions = tf.argmax(input=logits, axis=1)
       self.assertListEqual(predictions.get_shape().as_list(), [batch_size])
 
   def testTrainEvalWithReuse(self):
@@ -151,29 +151,29 @@ def testTrainEvalWithReuse(self):
     eval_height, eval_width = 300, 400
     num_classes = 1000
     with self.test_session():
-      train_inputs = tf.random_uniform(
+      train_inputs = tf.random.uniform(
           (train_batch_size, train_height, train_width, 3))
       logits, _ = alexnet.alexnet_v2(train_inputs)
       self.assertListEqual(logits.get_shape().as_list(),
                            [train_batch_size, num_classes])
-      tf.get_variable_scope().reuse_variables()
-      eval_inputs = tf.random_uniform(
+      tf.compat.v1.get_variable_scope().reuse_variables()
+      eval_inputs = tf.random.uniform(
           (eval_batch_size, eval_height, eval_width, 3))
       logits, _ = alexnet.alexnet_v2(eval_inputs, is_training=False,
                                      spatial_squeeze=False)
       self.assertListEqual(logits.get_shape().as_list(),
                            [eval_batch_size, 4, 7, num_classes])
-      logits = tf.reduce_mean(logits, [1, 2])
-      predictions = tf.argmax(logits, 1)
+      logits = tf.reduce_mean(input_tensor=logits, axis=[1, 2])
+      predictions = tf.argmax(input=logits, axis=1)
       self.assertEquals(predictions.get_shape().as_list(), [eval_batch_size])
 
   def testForward(self):
     batch_size = 1
     height, width = 224, 224
     with self.test_session() as sess:
-      inputs = tf.random_uniform((batch_size, height, width, 3))
+      inputs = tf.random.uniform((batch_size, height, width, 3))
       logits, _ = alexnet.alexnet_v2(inputs)
-      sess.run(tf.global_variables_initializer())
+      sess.run(tf.compat.v1.global_variables_initializer())
       output = sess.run(logits)
       self.assertTrue(output.any())
 

diff --git a/research/slim/nets/cifarnet.py b/research/slim/nets/cifarnet.py
@@ -23,7 +23,9 @@
 
 slim = contrib_slim
 
-trunc_normal = lambda stddev: tf.truncated_normal_initializer(stddev=stddev)
+# pylint: disable=g-long-lambda
+trunc_normal = lambda stddev: tf.compat.v1.truncated_normal_initializer(
+    stddev=stddev)
 
 
 def cifarnet(images, num_classes=10, is_training=False,
@@ -61,7 +63,7 @@ def cifarnet(images, num_classes=10, is_training=False,
   """
   end_points = {}
 
-  with tf.variable_scope(scope, 'CifarNet', [images]):
+  with tf.compat.v1.variable_scope(scope, 'CifarNet', [images]):
     net = slim.conv2d(images, 64, [5, 5], scope='conv1')
     end_points['conv1'] = net
     net = slim.max_pool2d(net, [2, 2], 2, scope='pool1')
@@ -82,12 +84,14 @@ def cifarnet(images, num_classes=10, is_training=False,
     end_points['fc4'] = net
     if not num_classes:
       return net, end_points
-    logits = slim.fully_connected(net, num_classes,
-                                  biases_initializer=tf.zeros_initializer(),
-                                  weights_initializer=trunc_normal(1/192.0),
-                                  weights_regularizer=None,
-                                  activation_fn=None,
-                                  scope='logits')
+    logits = slim.fully_connected(
+        net,
+        num_classes,
+        biases_initializer=tf.compat.v1.zeros_initializer(),
+        weights_initializer=trunc_normal(1 / 192.0),
+        weights_regularizer=None,
+        activation_fn=None,
+        scope='logits')
 
     end_points['Logits'] = logits
     end_points['Predictions'] = prediction_fn(logits, scope='Predictions')
@@ -107,11 +111,12 @@ def cifarnet_arg_scope(weight_decay=0.004):
   """
   with slim.arg_scope(
       [slim.conv2d],
-      weights_initializer=tf.truncated_normal_initializer(stddev=5e-2),
+      weights_initializer=tf.compat.v1.truncated_normal_initializer(
+          stddev=5e-2),
       activation_fn=tf.nn.relu):
     with slim.arg_scope(
         [slim.fully_connected],
-        biases_initializer=tf.constant_initializer(0.1),
+        biases_initializer=tf.compat.v1.constant_initializer(0.1),
         weights_initializer=trunc_normal(0.04),
         weights_regularizer=slim.l2_regularizer(weight_decay),
         activation_fn=tf.nn.relu) as sc: