Merged commit includes the following changes:

187187978  by Zhichao Lu:

    Only updating hyperparameters if they have non-null values.

--
187097690  by Zhichao Lu:

    Rewrite some conditions a bit more clearly.

--
187085190  by Zhichao Lu:

    More informative error message.

--
186935376  by Zhichao Lu:

    Added option to evaluator.evaluate to use custom evaluator objects.

--
186808249  by Zhichao Lu:

    Fix documentation re: number of stages.

--
186775014  by Zhichao Lu:

    Change anchor generator interface to return a list of BoxLists containing anchors for different feature map layers.

--
186729028  by Zhichao Lu:

    Minor fixes to object detection.

--
186723716  by Zhichao Lu:

    Fix tf_example_decoder.py initailization issue.

--
186668505  by Zhichao Lu:

    Remove unused import.

--
186475361  by Zhichao Lu:

    Update the box predictor interface to return list of predictions - one from each feature map - instead of stacking them into one large tensor.

--
186410844  by Zhichao Lu:

    Fix PythonPath Dependencies.

--
186365384  by Zhichao Lu:

    Made some of the functions in exporter public so they can be reused.

--
186341438  by Zhichao Lu:

    Re-introducing check that label-map-path must be a valid (non-empty) string prior to overwriting pipeline config.

--
186036984  by Zhichao Lu:

    Adding default hyperparameters and allowing for overriding them via flags.

--
186026006  by Zhichao Lu:

    Strip `eval_` prefix from name argument give to TPUEstimator.evaluate since it adds the same prefix internally.

--
186016042  by Zhichao Lu:

    Add an option to evaluate models on training data.

--
185944986  by Zhichao Lu:

    let _update_label_map_path go through even if the path is empty

--
185860781  by Zhichao Lu:

    Add random normal initializer option to hyperparams builder.

    Scale the regression losses outside of the box encoder by adjusting huber loss delta and regression loss weight.

--
185846325  by Zhichao Lu:

    Add an option to normalize localization loss by the code size(number of box coordinates) in SSD Meta architecture.

--
185761217  by Zhichao Lu:

    Change multiscale_grid_anchor_generator to return anchors in normalized coordinates by default and add option to configure it.

    In SSD meta architecture, TargetAssigner operates in normalized coordinate space (i.e, groundtruth boxes are in normalized coordinates) hence we need the option to generate anchors in normalized coordinates.

--
185747733  by Zhichao Lu:

    Change the smooth L1 localization implementationt to use tf.losses.huber_loss and expose the delta parameter in the proto.

--
185715309  by Zhichao Lu:

    Obviates the need for prepadding on mobilenet v1 and v2 for fully convolutional models.

--
185685695  by Zhichao Lu:

    Fix manual stepping schedule to return first rate when there are no boundaries

--
185621650  by Zhichao Lu:

    Added target assigner proto for configuring negative class weights.

--

PiperOrigin-RevId: 187187978

research/object_detection/models/feature_map_generators_test.py

@@ -40,7 +40,7 @@ EMBEDDED_SSD_MOBILENET_V1_LAYOUT = {
 }
 
 
-# TODO: add tests with different anchor strides.
+# TODO(rathodv): add tests with different anchor strides.
 class MultiResolutionFeatureMapGeneratorTest(tf.test.TestCase):
 
   def test_get_expected_feature_map_shapes_with_inception_v2(self):

research/object_detection/models/ssd_feature_extractor_test.py

@@ -27,13 +27,17 @@ from object_detection.utils import test_case
 class SsdFeatureExtractorTestBase(test_case.TestCase):
 
   @abstractmethod
-  def _create_feature_extractor(self, depth_multiplier, pad_to_multiple):
+  def _create_feature_extractor(self, depth_multiplier, pad_to_multiple,
+                                use_explicit_padding=False):
     """Constructs a new feature extractor.
 
     Args:
       depth_multiplier: float depth multiplier for feature extractor
       pad_to_multiple: the nearest multiple to zero pad the input height and
         width dimensions to.
+      use_explicit_padding: use 'VALID' padding for convolutions, but prepad
+        inputs so that the output dimensions are the same as if 'SAME' padding
+        were used.
     Returns:
       an ssd_meta_arch.SSDFeatureExtractor object.
     """
@@ -41,10 +45,11 @@ class SsdFeatureExtractorTestBase(test_case.TestCase):
 
   def check_extract_features_returns_correct_shape(
       self, batch_size, image_height, image_width, depth_multiplier,
-      pad_to_multiple, expected_feature_map_shapes):
+      pad_to_multiple, expected_feature_map_shapes, use_explicit_padding=False):
     def graph_fn(image_tensor):
       feature_extractor = self._create_feature_extractor(depth_multiplier,
-                                                         pad_to_multiple)
+                                                         pad_to_multiple,
+                                                         use_explicit_padding)
       feature_maps = feature_extractor.extract_features(image_tensor)
       return feature_maps
 
@@ -57,10 +62,11 @@ class SsdFeatureExtractorTestBase(test_case.TestCase):
 
   def check_extract_features_returns_correct_shapes_with_dynamic_inputs(
       self, batch_size, image_height, image_width, depth_multiplier,
-      pad_to_multiple, expected_feature_map_shapes):
+      pad_to_multiple, expected_feature_map_shapes, use_explicit_padding=False):
     def graph_fn(image_height, image_width):
       feature_extractor = self._create_feature_extractor(depth_multiplier,
-                                                         pad_to_multiple)
+                                                         pad_to_multiple,
+                                                         use_explicit_padding)
       image_tensor = tf.random_uniform([batch_size, image_height, image_width,
                                         3], dtype=tf.float32)
       feature_maps = feature_extractor.extract_features(image_tensor)

research/object_detection/models/ssd_mobilenet_v1_feature_extractor.py

@@ -53,8 +53,9 @@ class SSDMobileNetV1FeatureExtractor(ssd_meta_arch.SSDFeatureExtractor):
         (e.g. 1), it is desirable to disable batch norm update and use
         pretrained batch norm params.
       reuse_weights: Whether to reuse variables. Default is None.
-      use_explicit_padding: Whether to use explicit padding when extracting
-        features. Default is False.
+      use_explicit_padding: Use 'VALID' padding for convolutions, but prepad
+        inputs so that the output dimensions are the same as if 'SAME' padding
+        were used.
       use_depthwise: Whether to use depthwise convolutions. Default is False.
     """
     super(SSDMobileNetV1FeatureExtractor, self).__init__(
@@ -100,7 +101,7 @@ class SSDMobileNetV1FeatureExtractor(ssd_meta_arch.SSDFeatureExtractor):
     }
 
     with slim.arg_scope(self._conv_hyperparams):
-      # TODO: Enable fused batch norm once quantization supports it.
+      # TODO(skligys): Enable fused batch norm once quantization supports it.
       with slim.arg_scope([slim.batch_norm], fused=False):
         with tf.variable_scope('MobilenetV1',
                                reuse=self._reuse_weights) as scope:
@@ -109,6 +110,7 @@ class SSDMobileNetV1FeatureExtractor(ssd_meta_arch.SSDFeatureExtractor):
               final_endpoint='Conv2d_13_pointwise',
               min_depth=self._min_depth,
               depth_multiplier=self._depth_multiplier,
+              use_explicit_padding=self._use_explicit_padding,
               scope=scope)
           feature_maps = feature_map_generators.multi_resolution_feature_maps(
               feature_map_layout=feature_map_layout,

research/object_detection/models/ssd_mobilenet_v1_feature_extractor_test.py

@@ -27,7 +27,8 @@ class SsdMobilenetV1FeatureExtractorTest(
     ssd_feature_extractor_test.SsdFeatureExtractorTestBase):
 
   def _create_feature_extractor(self, depth_multiplier, pad_to_multiple,
-                                is_training=True, batch_norm_trainable=True):
+                                is_training=True, batch_norm_trainable=True,
+                                use_explicit_padding=False):
     """Constructs a new feature extractor.
 
     Args:
@@ -37,6 +38,9 @@ class SsdMobilenetV1FeatureExtractorTest(
       is_training: whether the network is in training mode.
       batch_norm_trainable: Whether to update batch norm parameters during
         training or not.
+      use_explicit_padding: Use 'VALID' padding for convolutions, but prepad
+        inputs so that the output dimensions are the same as if 'SAME' padding
+        were used.
     Returns:
       an ssd_meta_arch.SSDFeatureExtractor object.
     """
@@ -45,7 +49,8 @@ class SsdMobilenetV1FeatureExtractorTest(
       conv_hyperparams = sc
     return ssd_mobilenet_v1_feature_extractor.SSDMobileNetV1FeatureExtractor(
         is_training, depth_multiplier, min_depth, pad_to_multiple,
-        conv_hyperparams, batch_norm_trainable)
+        conv_hyperparams, batch_norm_trainable=batch_norm_trainable,
+        use_explicit_padding=use_explicit_padding)
 
   def test_extract_features_returns_correct_shapes_128(self):
     image_height = 128
@@ -57,7 +62,10 @@ class SsdMobilenetV1FeatureExtractorTest(
                                   (2, 1, 1, 256), (2, 1, 1, 128)]
     self.check_extract_features_returns_correct_shape(
         2, image_height, image_width, depth_multiplier, pad_to_multiple,
-        expected_feature_map_shape)
+        expected_feature_map_shape, use_explicit_padding=False)
+    self.check_extract_features_returns_correct_shape(
+        2, image_height, image_width, depth_multiplier, pad_to_multiple,
+        expected_feature_map_shape, use_explicit_padding=True)
 
   def test_extract_features_returns_correct_shapes_299(self):
     image_height = 299
@@ -69,7 +77,10 @@ class SsdMobilenetV1FeatureExtractorTest(
                                   (2, 2, 2, 256), (2, 1, 1, 128)]
     self.check_extract_features_returns_correct_shape(
         2, image_height, image_width, depth_multiplier, pad_to_multiple,
-        expected_feature_map_shape)
+        expected_feature_map_shape, use_explicit_padding=False)
+    self.check_extract_features_returns_correct_shape(
+        2, image_height, image_width, depth_multiplier, pad_to_multiple,
+        expected_feature_map_shape, use_explicit_padding=True)
 
   def test_extract_features_with_dynamic_image_shape(self):
     image_height = 128
@@ -81,7 +92,10 @@ class SsdMobilenetV1FeatureExtractorTest(
                                   (2, 1, 1, 256), (2, 1, 1, 128)]
     self.check_extract_features_returns_correct_shapes_with_dynamic_inputs(
         2, image_height, image_width, depth_multiplier, pad_to_multiple,
-        expected_feature_map_shape)
+        expected_feature_map_shape, use_explicit_padding=False)
+    self.check_extract_features_returns_correct_shape(
+        2, image_height, image_width, depth_multiplier, pad_to_multiple,
+        expected_feature_map_shape, use_explicit_padding=True)
 
   def test_extract_features_returns_correct_shapes_enforcing_min_depth(self):
     image_height = 299
@@ -93,7 +107,10 @@ class SsdMobilenetV1FeatureExtractorTest(
                                   (2, 2, 2, 32), (2, 1, 1, 32)]
     self.check_extract_features_returns_correct_shape(
         2, image_height, image_width, depth_multiplier, pad_to_multiple,
-        expected_feature_map_shape)
+        expected_feature_map_shape, use_explicit_padding=False)
+    self.check_extract_features_returns_correct_shape(
+        2, image_height, image_width, depth_multiplier, pad_to_multiple,
+        expected_feature_map_shape, use_explicit_padding=True)
 
   def test_extract_features_returns_correct_shapes_with_pad_to_multiple(self):
     image_height = 299
@@ -105,7 +122,10 @@ class SsdMobilenetV1FeatureExtractorTest(
                                   (2, 2, 2, 256), (2, 1, 1, 128)]
     self.check_extract_features_returns_correct_shape(
         2, image_height, image_width, depth_multiplier, pad_to_multiple,
-        expected_feature_map_shape)
+        expected_feature_map_shape, use_explicit_padding=False)
+    self.check_extract_features_returns_correct_shape(
+        2, image_height, image_width, depth_multiplier, pad_to_multiple,
+        expected_feature_map_shape, use_explicit_padding=True)
 
   def test_extract_features_raises_error_with_invalid_image_size(self):
     image_height = 32

research/object_detection/models/ssd_resnet_v1_fpn_feature_extractor.py

+# Copyright 2017 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.
+# ==============================================================================
 """SSD Feature Pyramid Network (FPN) feature extractors based on Resnet v1.
 
 See https://arxiv.org/abs/1708.02002 for details.
@@ -87,7 +101,7 @@ class _SSDResnetV1FpnFeatureExtractor(ssd_meta_arch.SSDFeatureExtractor):
     return resized_inputs - [[channel_means]]
 
   def _filter_features(self, image_features):
-    # TODO: Change resnet endpoint to strip scope prefixes instead
+    # TODO(rathodv): Change resnet endpoint to strip scope prefixes instead
     # of munging the scope here.
     filtered_image_features = dict({})
     for key, feature in image_features.items():

research/object_detection/models/ssd_resnet_v1_fpn_feature_extractor_test.py

+# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
 """Tests for ssd resnet v1 FPN feature extractors."""
 import tensorflow as tf
 
@@ -10,14 +24,16 @@ class SSDResnet50V1FeatureExtractorTest(
     SSDResnetFPNFeatureExtractorTestBase):
   """SSDResnet50v1Fpn feature extractor test."""
 
-  def _create_feature_extractor(self, depth_multiplier, pad_to_multiple):
+  def _create_feature_extractor(self, depth_multiplier, pad_to_multiple,
+                                use_explicit_padding=False):
     min_depth = 32
     conv_hyperparams = {}
     batch_norm_trainable = True
     is_training = True
     return ssd_resnet_v1_fpn_feature_extractor.SSDResnet50V1FpnFeatureExtractor(
         is_training, depth_multiplier, min_depth, pad_to_multiple,
-        conv_hyperparams, batch_norm_trainable)
+        conv_hyperparams, batch_norm_trainable,
+        use_explicit_padding=use_explicit_padding)
 
   def _resnet_scope_name(self):
     return 'resnet_v1_50'
@@ -28,7 +44,8 @@ class SSDResnet101V1FeatureExtractorTest(
     SSDResnetFPNFeatureExtractorTestBase):
   """SSDResnet101v1Fpn feature extractor test."""
 
-  def _create_feature_extractor(self, depth_multiplier, pad_to_multiple):
+  def _create_feature_extractor(self, depth_multiplier, pad_to_multiple,
+                                use_explicit_padding=False):
     min_depth = 32
     conv_hyperparams = {}
     batch_norm_trainable = True
@@ -36,7 +53,8 @@ class SSDResnet101V1FeatureExtractorTest(
     return (
         ssd_resnet_v1_fpn_feature_extractor.SSDResnet101V1FpnFeatureExtractor(
             is_training, depth_multiplier, min_depth, pad_to_multiple,
-            conv_hyperparams, batch_norm_trainable))
+            conv_hyperparams, batch_norm_trainable,
+            use_explicit_padding=use_explicit_padding))
 
   def _resnet_scope_name(self):
     return 'resnet_v1_101'
@@ -47,7 +65,8 @@ class SSDResnet152V1FeatureExtractorTest(
     SSDResnetFPNFeatureExtractorTestBase):
   """SSDResnet152v1Fpn feature extractor test."""
 
-  def _create_feature_extractor(self, depth_multiplier, pad_to_multiple):
+  def _create_feature_extractor(self, depth_multiplier, pad_to_multiple,
+                                use_explicit_padding=False):
     min_depth = 32
     conv_hyperparams = {}
     batch_norm_trainable = True
@@ -55,7 +74,8 @@ class SSDResnet152V1FeatureExtractorTest(
     return (
         ssd_resnet_v1_fpn_feature_extractor.SSDResnet152V1FpnFeatureExtractor(
             is_training, depth_multiplier, min_depth, pad_to_multiple,
-            conv_hyperparams, batch_norm_trainable))
+            conv_hyperparams, batch_norm_trainable,
+            use_explicit_padding=use_explicit_padding))
 
   def _resnet_scope_name(self):
     return 'resnet_v1_152'

research/object_detection/models/ssd_resnet_v1_fpn_feature_extractor_testbase.py

+# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
 """Tests for ssd resnet v1 FPN feature extractors."""
 import abc
 import numpy as np

research/object_detection/object_detection_tutorial.ipynb

@@ -58,7 +58,7 @@
    "outputs": [],
    "source": [
     "# This is needed to display the images.\n",
-    "%matplotlib inline"
+    "%matplotlib inline",
    ]
   },
   {

research/object_detection/protos/BUILD

-# Tensorflow Object Detection API: Configuration protos.
-
-package(
-    default_visibility = ["//visibility:public"],
-)
-
-licenses(["notice"])
-
-proto_library(
-    name = "argmax_matcher_proto",
-    srcs = ["argmax_matcher.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "argmax_matcher_py_pb2",
-    api_version = 2,
-    deps = [":argmax_matcher_proto"],
-)
-
-proto_library(
-    name = "bipartite_matcher_proto",
-    srcs = ["bipartite_matcher.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "bipartite_matcher_py_pb2",
-    api_version = 2,
-    deps = [":bipartite_matcher_proto"],
-)
-
-proto_library(
-    name = "matcher_proto",
-    srcs = ["matcher.proto"],
-    cc_api_version = 2,
-    deps = [
-        ":argmax_matcher_proto",
-        ":bipartite_matcher_proto",
-    ],
-)
-
-py_proto_library(
-    name = "matcher_py_pb2",
-    api_version = 2,
-    deps = [":matcher_proto"],
-)
-
-proto_library(
-    name = "faster_rcnn_box_coder_proto",
-    srcs = ["faster_rcnn_box_coder.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "faster_rcnn_box_coder_py_pb2",
-    api_version = 2,
-    deps = [":faster_rcnn_box_coder_proto"],
-)
-
-proto_library(
-    name = "keypoint_box_coder_proto",
-    srcs = ["keypoint_box_coder.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "keypoint_box_coder_py_pb2",
-    api_version = 2,
-    deps = [":keypoint_box_coder_proto"],
-)
-
-proto_library(
-    name = "mean_stddev_box_coder_proto",
-    srcs = ["mean_stddev_box_coder.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "mean_stddev_box_coder_py_pb2",
-    api_version = 2,
-    deps = [":mean_stddev_box_coder_proto"],
-)
-
-proto_library(
-    name = "square_box_coder_proto",
-    srcs = ["square_box_coder.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "square_box_coder_py_pb2",
-    api_version = 2,
-    deps = [":square_box_coder_proto"],
-)
-
-proto_library(
-    name = "box_coder_proto",
-    srcs = ["box_coder.proto"],
-    cc_api_version = 2,
-    deps = [
-        ":faster_rcnn_box_coder_proto",
-        ":keypoint_box_coder_proto",
-        ":mean_stddev_box_coder_proto",
-        ":square_box_coder_proto",
-    ],
-)
-
-py_proto_library(
-    name = "box_coder_py_pb2",
-    api_version = 2,
-    deps = [":box_coder_proto"],
-)
-
-proto_library(
-    name = "grid_anchor_generator_proto",
-    srcs = ["grid_anchor_generator.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "grid_anchor_generator_py_pb2",
-    api_version = 2,
-    deps = [":grid_anchor_generator_proto"],
-)
-
-proto_library(
-    name = "ssd_anchor_generator_proto",
-    srcs = ["ssd_anchor_generator.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "ssd_anchor_generator_py_pb2",
-    api_version = 2,
-    deps = [":ssd_anchor_generator_proto"],
-)
-
-proto_library(
-    name = "multiscale_anchor_generator_proto",
-    srcs = ["multiscale_anchor_generator.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "multiscale_anchor_generator_py_pb2",
-    api_version = 2,
-    deps = [":multiscale_anchor_generator_proto"],
-)
-
-proto_library(
-    name = "anchor_generator_proto",
-    srcs = ["anchor_generator.proto"],
-    cc_api_version = 2,
-    deps = [
-        ":grid_anchor_generator_proto",
-        ":multiscale_anchor_generator_proto",
-        ":ssd_anchor_generator_proto",
-    ],
-)
-
-py_proto_library(
-    name = "anchor_generator_py_pb2",
-    api_version = 2,
-    deps = [":anchor_generator_proto"],
-)
-
-proto_library(
-    name = "input_reader_proto",
-    srcs = ["input_reader.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "input_reader_py_pb2",
-    api_version = 2,
-    deps = [":input_reader_proto"],
-)
-
-proto_library(
-    name = "losses_proto",
-    srcs = ["losses.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "losses_py_pb2",
-    api_version = 2,
-    deps = [":losses_proto"],
-)
-
-proto_library(
-    name = "optimizer_proto",
-    srcs = ["optimizer.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "optimizer_py_pb2",
-    api_version = 2,
-    deps = [":optimizer_proto"],
-)
-
-proto_library(
-    name = "post_processing_proto",
-    srcs = ["post_processing.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "post_processing_py_pb2",
-    api_version = 2,
-    deps = [":post_processing_proto"],
-)
-
-proto_library(
-    name = "hyperparams_proto",
-    srcs = ["hyperparams.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "hyperparams_py_pb2",
-    api_version = 2,
-    deps = [":hyperparams_proto"],
-)
-
-proto_library(
-    name = "box_predictor_proto",
-    srcs = ["box_predictor.proto"],
-    cc_api_version = 2,
-    deps = [":hyperparams_proto"],
-)
-
-py_proto_library(
-    name = "box_predictor_py_pb2",
-    api_version = 2,
-    deps = [":box_predictor_proto"],
-)
-
-proto_library(
-    name = "region_similarity_calculator_proto",
-    srcs = ["region_similarity_calculator.proto"],
-    cc_api_version = 2,
-    deps = [],
-)
-
-py_proto_library(
-    name = "region_similarity_calculator_py_pb2",
-    api_version = 2,
-    deps = [":region_similarity_calculator_proto"],
-)
-
-proto_library(
-    name = "preprocessor_proto",
-    srcs = ["preprocessor.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "preprocessor_py_pb2",
-    api_version = 2,
-    deps = [":preprocessor_proto"],
-)
-
-proto_library(
-    name = "train_proto",
-    srcs = ["train.proto"],
-    cc_api_version = 2,
-    deps = [
-        ":optimizer_proto",
-        ":preprocessor_proto",
-    ],
-)
-
-py_proto_library(
-    name = "train_py_pb2",
-    api_version = 2,
-    deps = [":train_proto"],
-)
-
-proto_library(
-    name = "eval_proto",
-    srcs = ["eval.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "eval_py_pb2",
-    api_version = 2,
-    deps = [":eval_proto"],
-)
-
-proto_library(
-    name = "image_resizer_proto",
-    srcs = ["image_resizer.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "image_resizer_py_pb2",
-    api_version = 2,
-    deps = [":image_resizer_proto"],
-)
-
-proto_library(
-    name = "faster_rcnn_proto",
-    srcs = ["faster_rcnn.proto"],
-    cc_api_version = 2,
-    deps = [
-        ":box_predictor_proto",
-        "//tensorflow/models/research/object_detection/protos:anchor_generator_proto",
-        "//tensorflow/models/research/object_detection/protos:hyperparams_proto",
-        "//tensorflow/models/research/object_detection/protos:image_resizer_proto",
-        "//tensorflow/models/research/object_detection/protos:losses_proto",
-        "//tensorflow/models/research/object_detection/protos:post_processing_proto",
-    ],
-)
-
-proto_library(
-    name = "ssd_proto",
-    srcs = ["ssd.proto"],
-    cc_api_version = 2,
-    deps = [
-        ":anchor_generator_proto",
-        ":box_coder_proto",
-        ":box_predictor_proto",
-        ":hyperparams_proto",
-        ":image_resizer_proto",
-        ":losses_proto",
-        ":matcher_proto",
-        ":post_processing_proto",
-        ":region_similarity_calculator_proto",
-    ],
-)
-
-proto_library(
-    name = "model_proto",
-    srcs = ["model.proto"],
-    cc_api_version = 2,
-    deps = [
-        ":faster_rcnn_proto",
-        ":ssd_proto",
-    ],
-)
-
-py_proto_library(
-    name = "model_py_pb2",
-    api_version = 2,
-    deps = [":model_proto"],
-)
-
-proto_library(
-    name = "pipeline_proto",
-    srcs = ["pipeline.proto"],
-    cc_api_version = 2,
-    deps = [
-        ":eval_proto",
-        ":input_reader_proto",
-        ":model_proto",
-        ":train_proto",
-    ],
-)
-
-py_proto_library(
-    name = "pipeline_py_pb2",
-    api_version = 2,
-    deps = [":pipeline_proto"],
-)
-
-proto_library(
-    name = "string_int_label_map_proto",
-    srcs = ["string_int_label_map.proto"],
-    cc_api_version = 2,
-)
-
-py_proto_library(
-    name = "string_int_label_map_py_pb2",
-    api_version = 2,
-    deps = [":string_int_label_map_proto"],
-)

research/object_detection/protos/hyperparams.proto

@@ -65,6 +65,7 @@ message Initializer {
   oneof initializer_oneof {
     TruncatedNormalInitializer truncated_normal_initializer = 1;
     VarianceScalingInitializer variance_scaling_initializer = 2;
+    RandomNormalInitializer random_normal_initializer = 3;
   }
 }
 
@@ -89,6 +90,13 @@ message VarianceScalingInitializer {
   optional Mode mode = 3 [default = FAN_IN];
 }
 
+// Configuration proto for random normal initializer. See
+// https://www.tensorflow.org/api_docs/python/tf/random_normal_initializer
+message RandomNormalInitializer {
+  optional float mean = 1 [default = 0.0];
+  optional float stddev = 2 [default = 1.0];
+}
+
 // Configuration proto for batch norm to apply after convolution op. See
 // https://www.tensorflow.org/api_docs/python/tf/contrib/layers/batch_norm
 message BatchNorm {

research/object_detection/protos/losses.proto

@@ -38,11 +38,17 @@ message WeightedL2LocalizationLoss {
   optional bool anchorwise_output = 1 [default=false];
 }
 
-// SmoothL1 (Huber) location loss: .5 * x ^ 2 if |x| < 1 else |x| - .5
+// SmoothL1 (Huber) location loss.
+// The smooth L1_loss is defined elementwise as .5 x^2 if |x| <= delta and
+// 0.5 x^2 + delta * (|x|-delta) otherwise, where x is the difference between
+// predictions and target.
 message WeightedSmoothL1LocalizationLoss {
   // DEPRECATED, do not use.
   // Output loss per anchor.
   optional bool anchorwise_output = 1 [default=false];
+
+  // Delta value for huber loss.
+  optional float delta = 2 [default=1.0];
 }
 
 // Intersection over union location loss: 1 - IOU

research/object_detection/protos/multiscale_anchor_generator.proto

@@ -20,4 +20,7 @@ message MultiscaleAnchorGenerator {
 
   // Number of intermediate scale each scale octave
   optional int32 scales_per_octave = 5 [default = 2];
+
+  // Whether to produce anchors in normalized coordinates.
+  optional bool normalize_coordinates = 6 [default = true];
 }

research/object_detection/protos/ssd.proto

@@ -36,6 +36,10 @@ message Ssd {
   // zeros vector or a one-hot vector (where background is the 0th class).
   optional bool encode_background_as_zeros = 12 [default=false];
 
+  // classification weight to be associated to negative
+  // anchors (default: 1.0). The weight must be in [0., 1.].
+  optional float negative_class_weight = 13 [default = 1.0];
+
   // Box predictor to attach to the features.
   optional BoxPredictor box_predictor = 7;
 
@@ -49,6 +53,10 @@ message Ssd {
   // the anchors.
   optional bool normalize_loss_by_num_matches = 10 [default=true];
 
+  // Whether to normalize the localization loss by the code size of the box
+  // encodings. This is applied along with other normalization factors.
+  optional bool normalize_loc_loss_by_codesize = 14 [default=false];
+
   // Loss configuration for training.
   optional Loss loss = 11;
 }

research/object_detection/samples/configs/BUILD

-package(
-    default_visibility = ["//visibility:public"],
-)
-
-licenses(["notice"])
-
-exports_files([
-    "faster_rcnn_resnet50_pets.config",
-    "ssd_inception_v2_pets.config",
-    "ssd_mobilenet_v1_focal_loss_pets.config",
-])

research/object_detection/test_data/BUILD

-package(
-    default_visibility = ["//visibility:public"],
-)
-
-licenses(["notice"])
-
-exports_files([
-    "pets_examples.record",
-])

research/object_detection/test_images/BUILD

-package(
-    default_visibility = ["//visibility:public"],
-)
-
-licenses(["notice"])
-
-exports_files([
-    "image1.jpg",
-    "image2.jpg",
-])

research/object_detection/trainer.py

@@ -235,7 +235,7 @@ def train(create_tensor_dict_fn, create_model_fn, train_config, master, task,
           train_config.prefetch_queue_capacity, data_augmentation_options)
 
     # Gather initial summaries.
-    # TODO: See if summaries can be added/extracted from global tf
+    # TODO(rathodv): See if summaries can be added/extracted from global tf
     # collections so that they don't have to be passed around.
     summaries = set(tf.get_collection(tf.GraphKeys.SUMMARIES))
     global_summaries = set([])

research/object_detection/utils/config_util.py

@@ -241,6 +241,10 @@ def merge_external_params_with_configs(configs, hparams=None, **kwargs):
   if hparams:
     kwargs.update(hparams.values())
   for key, value in kwargs.items():
+    # pylint: disable=g-explicit-bool-comparison
+    if value == "" or value is None:
+      continue
+    # pylint: enable=g-explicit-bool-comparison
     if key == "learning_rate":
       _update_initial_learning_rate(configs, value)
       tf.logging.info("Overwriting learning rate: %f", value)
@@ -270,9 +274,8 @@ def merge_external_params_with_configs(configs, hparams=None, **kwargs):
       _update_input_path(configs["eval_input_config"], value)
       tf.logging.info("Overwriting eval input path: %s", value)
     if key == "label_map_path":
-      if value:
-        _update_label_map_path(configs, value)
-        tf.logging.info("Overwriting label map path: %s", value)
+      _update_label_map_path(configs, value)
+      tf.logging.info("Overwriting label map path: %s", value)
     if key == "mask_type":
       _update_mask_type(configs, value)
       tf.logging.info("Overwritten mask type: %s", value)

research/object_detection/utils/config_util_test.py

@@ -397,6 +397,27 @@ class ConfigUtilTest(tf.test.TestCase):
     self.assertEqual(new_label_map_path,
                      configs["eval_input_config"].label_map_path)
 
+  def testDontOverwriteEmptyLabelMapPath(self):
+    """Tests that label map path will not by overwritten with empty string."""
+    original_label_map_path = "path/to/original/label_map"
+    new_label_map_path = ""
+    pipeline_config_path = os.path.join(self.get_temp_dir(), "pipeline.config")
+
+    pipeline_config = pipeline_pb2.TrainEvalPipelineConfig()
+    train_input_reader = pipeline_config.train_input_reader
+    train_input_reader.label_map_path = original_label_map_path
+    eval_input_reader = pipeline_config.eval_input_reader
+    eval_input_reader.label_map_path = original_label_map_path
+    _write_config(pipeline_config, pipeline_config_path)
+
+    configs = config_util.get_configs_from_pipeline_file(pipeline_config_path)
+    configs = config_util.merge_external_params_with_configs(
+        configs, label_map_path=new_label_map_path)
+    self.assertEqual(original_label_map_path,
+                     configs["train_input_config"].label_map_path)
+    self.assertEqual(original_label_map_path,
+                     configs["eval_input_config"].label_map_path)
+
   def testNewMaskType(self):
     """Tests that mask type can be overwritten in input readers."""
     original_mask_type = input_reader_pb2.NUMERICAL_MASKS