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

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

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

--
311766063  by Sergio Guadarrama:

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

--
311624958  by Sergio Guadarrama:

    Updates README that doesn't render properly in github documentation

--
310980959  by Sergio Guadarrama:

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

--
310263156  by Sergio Guadarrama:

    Adds model breakdown for MobilenetV3

--
308640516  by Sergio Guadarrama:

    Internal change

308244396  by Sergio Guadarrama:

    GroupNormalization support for MobilenetV3.

--
307475800  by Sergio Guadarrama:

    Internal change

--
302077708  by Sergio Guadarrama:

    Remove `disable_tf2` behavior from slim py_library targets

--
301208453  by Sergio Guadarrama:

    Automated refactoring to make code Python 3 compatible.

--
300816672  by Sergio Guadarrama:

    Internal change

299433840  by Sergio Guadarrama:

    Internal change

299221609  by Sergio Guadarrama:

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

--
299179617  by Sergio Guadarrama:

    Internal change

299040784  by Sergio Guadarrama:

    Internal change

299036699  by Sergio Guadarrama:

    Internal change

298736510  by Sergio Guadarrama:

    Internal change

298732599  by Sergio Guadarrama:

    Internal change

298729507  by Sergio Guadarrama:

    Internal change

298253328  by Sergio Guadarrama:

    Internal change

297788346  by Sergio Guadarrama:

    Internal change

297785278  by Sergio Guadarrama:

    Internal change

297783127  by Sergio Guadarrama:

    Internal change

297725870  by Sergio Guadarrama:

    Internal change

297721811  by Sergio Guadarrama:

    Internal change

297711347  by Sergio Guadarrama:

    Internal change

297708059  by Sergio Guadarrama:

    Internal change

297701831  by Sergio Guadarrama:

    Internal change

297700038  by Sergio Guadarrama:

    Internal change

297670468  by Sergio Guadarrama:

    Internal change.

--
297350326  by Sergio Guadarrama:

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

--
297201668  by Sergio Guadarrama:

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

--
294483372  by Sergio Guadarrama:

    Internal change

PiperOrigin-RevId: 311933687

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

    Modify the LSTM feature extractors to be python 3 compatible.

--
311264357  by Menglong Zhu:

    Removes contrib.slim

--
308957207  by Menglong Zhu:

    Automated refactoring to make code Python 3 compatible.

--
306976470  by yongzhe:

    Internal change

306777559  by Menglong Zhu:

    Internal change

--
299232507  by lzyuan:

    Internal update.

--
299221735  by lzyuan:

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

--

PiperOrigin-RevId: 312578615

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

    Internal changes.

--

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

research/object_detection/models/ssd_inception_v2_feature_extractor_test.py

@@ -15,7 +15,7 @@
 
 """Tests for object_detection.models.ssd_inception_v2_feature_extractor."""
 import numpy as np
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.models import ssd_feature_extractor_test
 from object_detection.models import ssd_inception_v2_feature_extractor

research/object_detection/models/ssd_inception_v3_feature_extractor.py

@@ -15,8 +15,8 @@
 # ==============================================================================
 
 """SSDFeatureExtractor for InceptionV3 features."""
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
+import tf_slim as slim
 
 from object_detection.meta_architectures import ssd_meta_arch
 from object_detection.models import feature_map_generators
@@ -24,8 +24,6 @@ from object_detection.utils import ops
 from object_detection.utils import shape_utils
 from nets import inception_v3
 
-slim = contrib_slim
-
 
 class SSDInceptionV3FeatureExtractor(ssd_meta_arch.SSDFeatureExtractor):
   """SSD Feature Extractor using InceptionV3 features."""

research/object_detection/models/ssd_inception_v3_feature_extractor_test.py

@@ -15,7 +15,7 @@
 
 """Tests for object_detection.models.ssd_inception_v3_feature_extractor."""
 import numpy as np
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.models import ssd_feature_extractor_test
 from object_detection.models import ssd_inception_v3_feature_extractor

research/object_detection/models/ssd_mobiledet_feature_extractor.py

+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+"""SSDFeatureExtractor for MobileDet features."""
+
+import functools
+import numpy as np
+import tensorflow.compat.v1 as tf
+import tf_slim as slim
+
+from object_detection.meta_architectures import ssd_meta_arch
+from object_detection.models import feature_map_generators
+from object_detection.utils import ops
+from object_detection.utils import shape_utils
+
+
+BACKBONE_WEIGHT_DECAY = 4e-5
+
+
+def _scale_filters(filters, multiplier, base=8):
+  """Scale the filters accordingly to (multiplier, base)."""
+  round_half_up = int(int(filters) * multiplier / base + 0.5)
+  result = int(round_half_up * base)
+  return max(result, base)
+
+
+def _swish6(h):
+  with tf.name_scope('swish6'):
+    return h * tf.nn.relu6(h + np.float32(3)) * np.float32(1. / 6.)
+
+
+def _conv(h, filters, kernel_size, strides=1,
+          normalizer_fn=slim.batch_norm, activation_fn=tf.nn.relu6):
+  if activation_fn is None:
+    raise ValueError('Activation function cannot be None. Use tf.identity '
+                     'instead to better support quantized training.')
+  return slim.conv2d(
+      h,
+      filters,
+      kernel_size,
+      stride=strides,
+      activation_fn=activation_fn,
+      normalizer_fn=normalizer_fn,
+      weights_initializer=tf.initializers.he_normal(),
+      weights_regularizer=slim.l2_regularizer(BACKBONE_WEIGHT_DECAY),
+      padding='SAME')
+
+
+def _separable_conv(
+    h, filters, kernel_size, strides=1, activation_fn=tf.nn.relu6):
+  """Separable convolution layer."""
+  if activation_fn is None:
+    raise ValueError('Activation function cannot be None. Use tf.identity '
+                     'instead to better support quantized training.')
+  # Depthwise variant of He initialization derived under the principle proposed
+  # in the original paper. Note the original He normalization was designed for
+  # full convolutions and calling tf.initializers.he_normal() can over-estimate
+  # the fan-in of a depthwise kernel by orders of magnitude.
+  stddev = (2.0 / kernel_size**2)**0.5 / .87962566103423978
+  depthwise_initializer = tf.initializers.truncated_normal(stddev=stddev)
+  return slim.separable_conv2d(
+      h,
+      filters,
+      kernel_size,
+      stride=strides,
+      activation_fn=activation_fn,
+      normalizer_fn=slim.batch_norm,
+      weights_initializer=depthwise_initializer,
+      pointwise_initializer=tf.initializers.he_normal(),
+      weights_regularizer=slim.l2_regularizer(BACKBONE_WEIGHT_DECAY),
+      padding='SAME')
+
+
+def _squeeze_and_excite(h, hidden_dim, activation_fn=tf.nn.relu6):
+  with tf.variable_scope(None, default_name='SqueezeExcite'):
+    height, width = h.shape[1], h.shape[2]
+    u = slim.avg_pool2d(h, [height, width], stride=1, padding='VALID')
+    u = _conv(u, hidden_dim, 1,
+              normalizer_fn=None, activation_fn=activation_fn)
+    u = _conv(u, h.shape[-1], 1,
+              normalizer_fn=None, activation_fn=tf.nn.sigmoid)
+    return u * h
+
+
+def _inverted_bottleneck_no_expansion(
+    h, filters, activation_fn=tf.nn.relu6,
+    kernel_size=3, strides=1, use_se=False):
+  """Inverted bottleneck layer without the first 1x1 expansion convolution."""
+  with tf.variable_scope(None, default_name='IBNNoExpansion'):
+    # Setting filters to None will make _separable_conv a depthwise conv.
+    h = _separable_conv(
+        h, None, kernel_size, strides=strides, activation_fn=activation_fn)
+    if use_se:
+      hidden_dim = _scale_filters(h.shape[-1], 0.25)
+      h = _squeeze_and_excite(h, hidden_dim, activation_fn=activation_fn)
+    h = _conv(h, filters, 1, activation_fn=tf.identity)
+    return h
+
+
+def _inverted_bottleneck(
+    h, filters, activation_fn=tf.nn.relu6,
+    kernel_size=3, expansion=8, strides=1, use_se=False, residual=True):
+  """Inverted bottleneck layer."""
+  with tf.variable_scope(None, default_name='IBN'):
+    shortcut = h
+    expanded_filters = int(h.shape[-1]) * expansion
+    if expansion <= 1:
+      raise ValueError('Expansion factor must be greater than 1.')
+    h = _conv(h, expanded_filters, 1, activation_fn=activation_fn)
+    # Setting filters to None will make _separable_conv a depthwise conv.
+    h = _separable_conv(h, None, kernel_size, strides=strides,
+                        activation_fn=activation_fn)
+    if use_se:
+      hidden_dim = _scale_filters(expanded_filters, 0.25)
+      h = _squeeze_and_excite(h, hidden_dim, activation_fn=activation_fn)
+    h = _conv(h, filters, 1, activation_fn=tf.identity)
+    if residual:
+      h = h + shortcut
+    return h
+
+
+def _fused_conv(
+    h, filters, activation_fn=tf.nn.relu6,
+    kernel_size=3, expansion=8, strides=1, use_se=False, residual=True):
+  """Fused convolution layer."""
+  with tf.variable_scope(None, default_name='FusedConv'):
+    shortcut = h
+    expanded_filters = int(h.shape[-1]) * expansion
+    if expansion <= 1:
+      raise ValueError('Expansion factor must be greater than 1.')
+    h = _conv(h, expanded_filters, kernel_size, strides=strides,
+              activation_fn=activation_fn)
+    if use_se:
+      hidden_dim = _scale_filters(expanded_filters, 0.25)
+      h = _squeeze_and_excite(h, hidden_dim, activation_fn=activation_fn)
+    h = _conv(h, filters, 1, activation_fn=tf.identity)
+    if residual:
+      h = h + shortcut
+    return h
+
+
+def _tucker_conv(
+    h, filters, activation_fn=tf.nn.relu6,
+    kernel_size=3, input_rank_ratio=0.25, output_rank_ratio=0.25,
+    strides=1, residual=True):
+  """Tucker convolution layer (generalized bottleneck)."""
+  with tf.variable_scope(None, default_name='TuckerConv'):
+    shortcut = h
+    input_rank = _scale_filters(h.shape[-1], input_rank_ratio)
+    h = _conv(h, input_rank, 1, activation_fn=activation_fn)
+    output_rank = _scale_filters(filters, output_rank_ratio)
+    h = _conv(h, output_rank, kernel_size, strides=strides,
+              activation_fn=activation_fn)
+    h = _conv(h, filters, 1, activation_fn=tf.identity)
+    if residual:
+      h = h + shortcut
+    return h
+
+
+def mobiledet_cpu_backbone(h, multiplier=1.0):
+  """Build a MobileDet CPU backbone."""
+  def _scale(filters):
+    return _scale_filters(filters, multiplier)
+  ibn = functools.partial(
+      _inverted_bottleneck, use_se=True, activation_fn=_swish6)
+
+  endpoints = {}
+  h = _conv(h, _scale(16), 3, strides=2, activation_fn=_swish6)
+  h = _inverted_bottleneck_no_expansion(
+      h, _scale(8), use_se=True, activation_fn=_swish6)
+  endpoints['C1'] = h
+  h = ibn(h, _scale(16), expansion=4, strides=2, residual=False)
+  endpoints['C2'] = h
+  h = ibn(h, _scale(32), expansion=8, strides=2, residual=False)
+  h = ibn(h, _scale(32), expansion=4)
+  h = ibn(h, _scale(32), expansion=4)
+  h = ibn(h, _scale(32), expansion=4)
+  endpoints['C3'] = h
+  h = ibn(h, _scale(72), kernel_size=5, expansion=8, strides=2, residual=False)
+  h = ibn(h, _scale(72), expansion=8)
+  h = ibn(h, _scale(72), kernel_size=5, expansion=4)
+  h = ibn(h, _scale(72), expansion=4)
+  h = ibn(h, _scale(72), expansion=8, residual=False)
+  h = ibn(h, _scale(72), expansion=8)
+  h = ibn(h, _scale(72), expansion=8)
+  h = ibn(h, _scale(72), expansion=8)
+  endpoints['C4'] = h
+  h = ibn(h, _scale(104), kernel_size=5, expansion=8, strides=2, residual=False)
+  h = ibn(h, _scale(104), kernel_size=5, expansion=4)
+  h = ibn(h, _scale(104), kernel_size=5, expansion=4)
+  h = ibn(h, _scale(104), expansion=4)
+  h = ibn(h, _scale(144), expansion=8, residual=False)
+  endpoints['C5'] = h
+  return endpoints
+
+
+def mobiledet_dsp_backbone(h, multiplier=1.0):
+  """Build a MobileDet DSP backbone."""
+  def _scale(filters):
+    return _scale_filters(filters, multiplier)
+
+  ibn = functools.partial(_inverted_bottleneck, activation_fn=tf.nn.relu6)
+  fused = functools.partial(_fused_conv, activation_fn=tf.nn.relu6)
+  tucker = functools.partial(_tucker_conv, activation_fn=tf.nn.relu6)
+
+  endpoints = {}
+  h = _conv(h, _scale(32), 3, strides=2, activation_fn=tf.nn.relu6)
+  h = _inverted_bottleneck_no_expansion(
+      h, _scale(24), activation_fn=tf.nn.relu6)
+  endpoints['C1'] = h
+  h = fused(h, _scale(32), expansion=4, strides=2, residual=False)
+  h = fused(h, _scale(32), expansion=4)
+  h = ibn(h, _scale(32), expansion=4)
+  h = tucker(h, _scale(32), input_rank_ratio=0.25, output_rank_ratio=0.75)
+  endpoints['C2'] = h
+  h = fused(h, _scale(64), expansion=8, strides=2, residual=False)
+  h = ibn(h, _scale(64), expansion=4)
+  h = fused(h, _scale(64), expansion=4)
+  h = fused(h, _scale(64), expansion=4)
+  endpoints['C3'] = h
+  h = fused(h, _scale(120), expansion=8, strides=2, residual=False)
+  h = ibn(h, _scale(120), expansion=4)
+  h = ibn(h, _scale(120), expansion=8)
+  h = ibn(h, _scale(120), expansion=8)
+  h = fused(h, _scale(144), expansion=8, residual=False)
+  h = ibn(h, _scale(144), expansion=8)
+  h = ibn(h, _scale(144), expansion=8)
+  h = ibn(h, _scale(144), expansion=8)
+  endpoints['C4'] = h
+  h = ibn(h, _scale(160), expansion=4, strides=2, residual=False)
+  h = ibn(h, _scale(160), expansion=4)
+  h = fused(h, _scale(160), expansion=4)
+  h = tucker(h, _scale(160), input_rank_ratio=0.75, output_rank_ratio=0.75)
+  h = ibn(h, _scale(240), expansion=8, residual=False)
+  endpoints['C5'] = h
+  return endpoints
+
+
+def mobiledet_edgetpu_backbone(h, multiplier=1.0):
+  """Build a MobileDet EdgeTPU backbone."""
+  def _scale(filters):
+    return _scale_filters(filters, multiplier)
+
+  ibn = functools.partial(_inverted_bottleneck, activation_fn=tf.nn.relu6)
+  fused = functools.partial(_fused_conv, activation_fn=tf.nn.relu6)
+  tucker = functools.partial(_tucker_conv, activation_fn=tf.nn.relu6)
+
+  endpoints = {}
+  h = _conv(h, _scale(32), 3, strides=2, activation_fn=tf.nn.relu6)
+  h = tucker(h, _scale(16),
+             input_rank_ratio=0.25, output_rank_ratio=0.75, residual=False)
+  endpoints['C1'] = h
+  h = fused(h, _scale(16), expansion=8, strides=2, residual=False)
+  h = fused(h, _scale(16), expansion=4)
+  h = fused(h, _scale(16), expansion=8)
+  h = fused(h, _scale(16), expansion=4)
+  endpoints['C2'] = h
+  h = fused(h, _scale(40), expansion=8, kernel_size=5, strides=2,
+            residual=False)
+  h = fused(h, _scale(40), expansion=4)
+  h = fused(h, _scale(40), expansion=4)
+  h = fused(h, _scale(40), expansion=4)
+  endpoints['C3'] = h
+  h = ibn(h, _scale(72), expansion=8, strides=2, residual=False)
+  h = ibn(h, _scale(72), expansion=8)
+  h = fused(h, _scale(72), expansion=4)
+  h = fused(h, _scale(72), expansion=4)
+  h = ibn(h, _scale(96), expansion=8, kernel_size=5, residual=False)
+  h = ibn(h, _scale(96), expansion=8, kernel_size=5)
+  h = ibn(h, _scale(96), expansion=8)
+  h = ibn(h, _scale(96), expansion=8)
+  endpoints['C4'] = h
+  h = ibn(h, _scale(120), expansion=8, kernel_size=5, strides=2, residual=False)
+  h = ibn(h, _scale(120), expansion=8)
+  h = ibn(h, _scale(120), expansion=4, kernel_size=5)
+  h = ibn(h, _scale(120), expansion=8)
+  h = ibn(h, _scale(384), expansion=8, kernel_size=5, residual=False)
+  endpoints['C5'] = h
+  return endpoints
+
+
+class SSDMobileDetFeatureExtractorBase(ssd_meta_arch.SSDFeatureExtractor):
+  """Base class of SSD feature extractor using MobileDet features."""
+
+  def __init__(self,
+               backbone_fn,
+               is_training,
+               depth_multiplier,
+               min_depth,
+               pad_to_multiple,
+               conv_hyperparams_fn,
+               reuse_weights=None,
+               use_explicit_padding=False,
+               use_depthwise=False,
+               override_base_feature_extractor_hyperparams=False,
+               scope_name='MobileDet'):
+    """MobileDet Feature Extractor for SSD Models.
+
+    Reference:
+      https://arxiv.org/abs/2004.14525
+
+    Args:
+      backbone_fn: function to construct the MobileDet backbone.
+      is_training: whether the network is in training mode.
+      depth_multiplier: float depth multiplier for feature extractor.
+      min_depth: Integer, minimum feature extractor depth (number of filters).
+      pad_to_multiple: the nearest multiple to zero pad the input height and
+        width dimensions to.
+      conv_hyperparams_fn: A function to construct tf slim arg_scope for conv2d
+        and separable_conv2d ops in the layers that are added on top of the base
+        feature extractor.
+      reuse_weights: Whether to reuse variables. Default is None.
+      use_explicit_padding: Whether to use explicit padding when extracting
+        features.
+      use_depthwise: Whether to use depthwise convolutions in the SSD head.
+      override_base_feature_extractor_hyperparams: Whether to override
+        hyperparameters of the base feature extractor with the one from
+        `conv_hyperparams_fn`.
+      scope_name: scope name (string) of network variables.
+    """
+    if use_explicit_padding:
+      raise NotImplementedError(
+          'Explicit padding is not yet supported in MobileDet backbones.')
+
+    super(SSDMobileDetFeatureExtractorBase, self).__init__(
+        is_training=is_training,
+        depth_multiplier=depth_multiplier,
+        min_depth=min_depth,
+        pad_to_multiple=pad_to_multiple,
+        conv_hyperparams_fn=conv_hyperparams_fn,
+        reuse_weights=reuse_weights,
+        use_explicit_padding=use_explicit_padding,
+        use_depthwise=use_depthwise,
+        override_base_feature_extractor_hyperparams=override_base_feature_extractor_hyperparams
+    )
+    self._backbone_fn = backbone_fn
+    self._scope_name = scope_name
+
+  def preprocess(self, resized_inputs):
+    """SSD preprocessing.
+
+    Maps pixel values to the range [-1, 1]. The preprocessing assumes an input
+    value range of [0, 255].
+
+    Args:
+      resized_inputs: a [batch, height, width, channels] float tensor
+        representing a batch of images.
+
+    Returns:
+      preprocessed_inputs: a [batch, height, width, channels] float tensor
+        representing a batch of images.
+    """
+    return (2.0 / 255.0) * resized_inputs - 1.0
+
+  def extract_features(self, preprocessed_inputs):
+    """Extract features from preprocessed inputs.
+
+    Args:
+      preprocessed_inputs: a [batch, height, width, channels] float tensor
+        representing a batch of images.
+
+    Returns:
+      feature_maps: a list of tensors where the ith tensor has shape
+        [batch, height_i, width_i, depth_i]
+    """
+    preprocessed_inputs = shape_utils.check_min_image_dim(
+        33, preprocessed_inputs)
+    padded_inputs = ops.pad_to_multiple(
+        preprocessed_inputs, self._pad_to_multiple)
+
+    feature_map_layout = {
+        'from_layer': ['C4', 'C5', '', '', '', ''],
+        # Do not specify the layer depths (number of filters) for C4 and C5, as
+        # their values are determined based on the backbone.
+        'layer_depth': [-1, -1, 512, 256, 256, 128],
+        'use_depthwise': self._use_depthwise,
+        'use_explicit_padding': self._use_explicit_padding,
+    }
+
+    with tf.variable_scope(self._scope_name, reuse=self._reuse_weights):
+      with slim.arg_scope([slim.batch_norm],
+                          is_training=self._is_training,
+                          epsilon=0.01, decay=0.99, center=True, scale=True):
+        endpoints = self._backbone_fn(
+            padded_inputs,
+            multiplier=self._depth_multiplier)
+
+      image_features = {'C4': endpoints['C4'], 'C5': endpoints['C5']}
+      with slim.arg_scope(self._conv_hyperparams_fn()):
+        feature_maps = feature_map_generators.multi_resolution_feature_maps(
+            feature_map_layout=feature_map_layout,
+            depth_multiplier=self._depth_multiplier,
+            min_depth=self._min_depth,
+            insert_1x1_conv=True,
+            image_features=image_features)
+
+    return list(feature_maps.values())
+
+
+class SSDMobileDetCPUFeatureExtractor(SSDMobileDetFeatureExtractorBase):
+  """MobileDet-CPU feature extractor."""
+
+  def __init__(self,
+               is_training,
+               depth_multiplier,
+               min_depth,
+               pad_to_multiple,
+               conv_hyperparams_fn,
+               reuse_weights=None,
+               use_explicit_padding=False,
+               use_depthwise=False,
+               override_base_feature_extractor_hyperparams=False,
+               scope_name='MobileDetCPU'):
+    super(SSDMobileDetCPUFeatureExtractor, self).__init__(
+        backbone_fn=mobiledet_cpu_backbone,
+        is_training=is_training,
+        depth_multiplier=depth_multiplier,
+        min_depth=min_depth,
+        pad_to_multiple=pad_to_multiple,
+        conv_hyperparams_fn=conv_hyperparams_fn,
+        reuse_weights=reuse_weights,
+        use_explicit_padding=use_explicit_padding,
+        use_depthwise=use_depthwise,
+        override_base_feature_extractor_hyperparams=override_base_feature_extractor_hyperparams,
+        scope_name=scope_name)
+
+
+class SSDMobileDetDSPFeatureExtractor(SSDMobileDetFeatureExtractorBase):
+  """MobileDet-DSP feature extractor."""
+
+  def __init__(self,
+               is_training,
+               depth_multiplier,
+               min_depth,
+               pad_to_multiple,
+               conv_hyperparams_fn,
+               reuse_weights=None,
+               use_explicit_padding=False,
+               use_depthwise=False,
+               override_base_feature_extractor_hyperparams=False,
+               scope_name='MobileDetDSP'):
+    super(SSDMobileDetDSPFeatureExtractor, self).__init__(
+        backbone_fn=mobiledet_dsp_backbone,
+        is_training=is_training,
+        depth_multiplier=depth_multiplier,
+        min_depth=min_depth,
+        pad_to_multiple=pad_to_multiple,
+        conv_hyperparams_fn=conv_hyperparams_fn,
+        reuse_weights=reuse_weights,
+        use_explicit_padding=use_explicit_padding,
+        use_depthwise=use_depthwise,
+        override_base_feature_extractor_hyperparams=override_base_feature_extractor_hyperparams,
+        scope_name=scope_name)
+
+
+class SSDMobileDetEdgeTPUFeatureExtractor(SSDMobileDetFeatureExtractorBase):
+  """MobileDet-EdgeTPU feature extractor."""
+
+  def __init__(self,
+               is_training,
+               depth_multiplier,
+               min_depth,
+               pad_to_multiple,
+               conv_hyperparams_fn,
+               reuse_weights=None,
+               use_explicit_padding=False,
+               use_depthwise=False,
+               override_base_feature_extractor_hyperparams=False,
+               scope_name='MobileDetEdgeTPU'):
+    super(SSDMobileDetEdgeTPUFeatureExtractor, self).__init__(
+        backbone_fn=mobiledet_edgetpu_backbone,
+        is_training=is_training,
+        depth_multiplier=depth_multiplier,
+        min_depth=min_depth,
+        pad_to_multiple=pad_to_multiple,
+        conv_hyperparams_fn=conv_hyperparams_fn,
+        reuse_weights=reuse_weights,
+        use_explicit_padding=use_explicit_padding,
+        use_depthwise=use_depthwise,
+        override_base_feature_extractor_hyperparams=override_base_feature_extractor_hyperparams,
+        scope_name=scope_name)

research/object_detection/models/ssd_mobiledet_feature_extractor_test.py

+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+"""Tests for ssd_mobiledet_feature_extractor."""
+
+import tensorflow.compat.v1 as tf
+
+from tensorflow.contrib import quantize as contrib_quantize
+from object_detection.models import ssd_feature_extractor_test
+from object_detection.models import ssd_mobiledet_feature_extractor
+
+
+class SSDMobileDetFeatureExtractorTest(
+    ssd_feature_extractor_test.SsdFeatureExtractorTestBase):
+
+  def _create_feature_extractor(self,
+                                feature_extractor_cls,
+                                is_training=False,
+                                depth_multiplier=1.0,
+                                pad_to_multiple=1,
+                                use_explicit_padding=False,
+                                use_keras=False):
+    """Constructs a new MobileDet feature extractor.
+
+    Args:
+      feature_extractor_cls: feature extractor class.
+      is_training: whether the network is in training mode.
+      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: If True, we will use 'VALID' padding for
+        convolutions, but prepad inputs so that the output dimensions are the
+        same as if 'SAME' padding were used.
+      use_keras: if True builds a keras-based feature extractor, if False builds
+        a slim-based one.
+
+    Returns:
+      an ssd_meta_arch.SSDMobileDetFeatureExtractor object.
+    """
+    min_depth = 32
+    return feature_extractor_cls(
+        is_training,
+        depth_multiplier,
+        min_depth,
+        pad_to_multiple,
+        self.conv_hyperparams_fn,
+        use_explicit_padding=use_explicit_padding)
+
+  def test_mobiledet_cpu_returns_correct_shapes(self):
+    expected_feature_map_shapes = [(2, 40, 20, 72),
+                                   (2, 20, 10, 144),
+                                   (2, 10, 5, 512),
+                                   (2, 5, 3, 256),
+                                   (2, 3, 2, 256),
+                                   (2, 2, 1, 128)]
+    feature_extractor = self._create_feature_extractor(
+        ssd_mobiledet_feature_extractor.SSDMobileDetCPUFeatureExtractor)
+    image = tf.random.normal((2, 640, 320, 3))
+    feature_maps = feature_extractor.extract_features(image)
+
+    self.assertEqual(len(expected_feature_map_shapes), len(feature_maps))
+    for expected_shape, x in zip(expected_feature_map_shapes, feature_maps):
+      self.assertTrue(x.shape.is_compatible_with(expected_shape))
+
+  def test_mobiledet_dsp_returns_correct_shapes(self):
+    expected_feature_map_shapes = [(2, 40, 20, 144),
+                                   (2, 20, 10, 240),
+                                   (2, 10, 5, 512),
+                                   (2, 5, 3, 256),
+                                   (2, 3, 2, 256),
+                                   (2, 2, 1, 128)]
+    feature_extractor = self._create_feature_extractor(
+        ssd_mobiledet_feature_extractor.SSDMobileDetDSPFeatureExtractor)
+    image = tf.random.normal((2, 640, 320, 3))
+    feature_maps = feature_extractor.extract_features(image)
+
+    self.assertEqual(len(expected_feature_map_shapes), len(feature_maps))
+    for expected_shape, x in zip(expected_feature_map_shapes, feature_maps):
+      self.assertTrue(x.shape.is_compatible_with(expected_shape))
+
+  def test_mobiledet_edgetpu_returns_correct_shapes(self):
+    expected_feature_map_shapes = [(2, 40, 20, 96),
+                                   (2, 20, 10, 384),
+                                   (2, 10, 5, 512),
+                                   (2, 5, 3, 256),
+                                   (2, 3, 2, 256),
+                                   (2, 2, 1, 128)]
+    feature_extractor = self._create_feature_extractor(
+        ssd_mobiledet_feature_extractor.SSDMobileDetEdgeTPUFeatureExtractor)
+    image = tf.random.normal((2, 640, 320, 3))
+    feature_maps = feature_extractor.extract_features(image)
+
+    self.assertEqual(len(expected_feature_map_shapes), len(feature_maps))
+    for expected_shape, x in zip(expected_feature_map_shapes, feature_maps):
+      self.assertTrue(x.shape.is_compatible_with(expected_shape))
+
+  def _check_quantization(self, model_fn):
+    checkpoint_dir = self.get_temp_dir()
+
+    with tf.Graph().as_default() as training_graph:
+      model_fn(is_training=True)
+      contrib_quantize.experimental_create_training_graph(training_graph)
+      with self.session(graph=training_graph) as sess:
+        sess.run(tf.global_variables_initializer())
+        tf.train.Saver().save(sess, checkpoint_dir)
+
+    with tf.Graph().as_default() as eval_graph:
+      model_fn(is_training=False)
+      contrib_quantize.experimental_create_eval_graph(eval_graph)
+      with self.session(graph=eval_graph) as sess:
+        tf.train.Saver().restore(sess, checkpoint_dir)
+
+  def test_mobiledet_cpu_quantization(self):
+    def model_fn(is_training):
+      feature_extractor = self._create_feature_extractor(
+          ssd_mobiledet_feature_extractor.SSDMobileDetCPUFeatureExtractor,
+          is_training=is_training)
+      image = tf.random.normal((2, 320, 320, 3))
+      feature_extractor.extract_features(image)
+    self._check_quantization(model_fn)
+
+  def test_mobiledet_dsp_quantization(self):
+    def model_fn(is_training):
+      feature_extractor = self._create_feature_extractor(
+          ssd_mobiledet_feature_extractor.SSDMobileDetDSPFeatureExtractor,
+          is_training=is_training)
+      image = tf.random.normal((2, 320, 320, 3))
+      feature_extractor.extract_features(image)
+    self._check_quantization(model_fn)
+
+  def test_mobiledet_edgetpu_quantization(self):
+    def model_fn(is_training):
+      feature_extractor = self._create_feature_extractor(
+          ssd_mobiledet_feature_extractor.SSDMobileDetEdgeTPUFeatureExtractor,
+          is_training=is_training)
+      image = tf.random.normal((2, 320, 320, 3))
+      feature_extractor.extract_features(image)
+    self._check_quantization(model_fn)
+
+
+if __name__ == '__main__':
+  tf.test.main()

research/object_detection/models/ssd_mobilenet_edgetpu_feature_extractor_test.py

@@ -14,14 +14,11 @@
 # ==============================================================================
 """Tests for ssd_mobilenet_edgetpu_feature_extractor."""
 
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
-from tensorflow.contrib import slim as contrib_slim
 from object_detection.models import ssd_mobilenet_edgetpu_feature_extractor
 from object_detection.models import ssd_mobilenet_edgetpu_feature_extractor_testbase
 
-slim = contrib_slim
-
 
 class SsdMobilenetEdgeTPUFeatureExtractorTest(
     ssd_mobilenet_edgetpu_feature_extractor_testbase

research/object_detection/models/ssd_mobilenet_edgetpu_feature_extractor_testbase.py

@@ -17,7 +17,7 @@
 import abc
 
 import numpy as np
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.models import ssd_feature_extractor_test
 

research/object_detection/models/ssd_mobilenet_v1_feature_extractor.py

@@ -16,8 +16,8 @@
 
 """SSDFeatureExtractor for MobilenetV1 features."""
 
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
+import tf_slim as slim
 
 from object_detection.meta_architectures import ssd_meta_arch
 from object_detection.models import feature_map_generators
@@ -26,8 +26,6 @@ from object_detection.utils import ops
 from object_detection.utils import shape_utils
 from nets import mobilenet_v1
 
-slim = contrib_slim
-
 
 class SSDMobileNetV1FeatureExtractor(ssd_meta_arch.SSDFeatureExtractor):
   """SSD Feature Extractor using MobilenetV1 features."""

research/object_detection/models/ssd_mobilenet_v1_feature_extractor_test.py

@@ -20,15 +20,12 @@ Keras-based Mobilenet V1 feature extractors in SSD.
 from absl.testing import parameterized
 
 import numpy as np
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
 
 from object_detection.models import ssd_feature_extractor_test
 from object_detection.models import ssd_mobilenet_v1_feature_extractor
 from object_detection.models import ssd_mobilenet_v1_keras_feature_extractor
 
-slim = contrib_slim
-
 
 @parameterized.parameters(
     {'use_keras': False},

research/object_detection/models/ssd_mobilenet_v1_fpn_feature_extractor.py

@@ -23,8 +23,8 @@ from __future__ import print_function
 import copy
 import functools
 from six.moves import range
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
+import tf_slim as slim
 
 from object_detection.meta_architectures import ssd_meta_arch
 from object_detection.models import feature_map_generators
@@ -33,8 +33,6 @@ from object_detection.utils import ops
 from object_detection.utils import shape_utils
 from nets import mobilenet_v1
 
-slim = contrib_slim
-
 
 # A modified config of mobilenet v1 that makes it more detection friendly,
 def _create_modified_mobilenet_config():

research/object_detection/models/ssd_mobilenet_v1_fpn_feature_extractor_test.py

@@ -20,15 +20,12 @@ Keras-based Mobilenet V1 FPN feature extractors in SSD.
 """
 from absl.testing import parameterized
 import numpy as np
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
 
 from object_detection.models import ssd_feature_extractor_test
 from object_detection.models import ssd_mobilenet_v1_fpn_feature_extractor
 from object_detection.models import ssd_mobilenet_v1_fpn_keras_feature_extractor
 
-slim = contrib_slim
-
 
 @parameterized.parameters(
     {'use_keras': False},

research/object_detection/models/ssd_mobilenet_v1_fpn_keras_feature_extractor.py

@@ -21,7 +21,7 @@ from __future__ import division
 from __future__ import print_function
 
 from six.moves import range
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.meta_architectures import ssd_meta_arch
 from object_detection.models import feature_map_generators

research/object_detection/models/ssd_mobilenet_v1_keras_feature_extractor.py

@@ -16,8 +16,7 @@
 
 """SSDFeatureExtractor for Keras MobilenetV1 features."""
 
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
 
 from object_detection.meta_architectures import ssd_meta_arch
 from object_detection.models import feature_map_generators
@@ -25,8 +24,6 @@ from object_detection.models.keras_models import mobilenet_v1
 from object_detection.utils import ops
 from object_detection.utils import shape_utils
 
-slim = contrib_slim
-
 
 class SSDMobileNetV1KerasFeatureExtractor(
     ssd_meta_arch.SSDKerasFeatureExtractor):

research/object_detection/models/ssd_mobilenet_v1_ppn_feature_extractor.py

@@ -16,8 +16,8 @@
 
 """SSDFeatureExtractor for MobilenetV1 PPN features."""
 
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
+import tf_slim as slim
 
 from object_detection.meta_architectures import ssd_meta_arch
 from object_detection.models import feature_map_generators
@@ -26,8 +26,6 @@ from object_detection.utils import ops
 from object_detection.utils import shape_utils
 from nets import mobilenet_v1
 
-slim = contrib_slim
-
 
 class SSDMobileNetV1PpnFeatureExtractor(ssd_meta_arch.SSDFeatureExtractor):
   """SSD Feature Extractor using MobilenetV1 PPN features."""

research/object_detection/models/ssd_mobilenet_v1_ppn_feature_extractor_test.py

@@ -15,14 +15,11 @@
 
 """Tests for ssd_mobilenet_v1_ppn_feature_extractor."""
 import numpy as np
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
 
 from object_detection.models import ssd_feature_extractor_test
 from object_detection.models import ssd_mobilenet_v1_ppn_feature_extractor
 
-slim = contrib_slim
-
 
 class SsdMobilenetV1PpnFeatureExtractorTest(
     ssd_feature_extractor_test.SsdFeatureExtractorTestBase):

research/object_detection/models/ssd_mobilenet_v2_feature_extractor.py

@@ -16,8 +16,8 @@
 
 """SSDFeatureExtractor for MobilenetV2 features."""
 
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
+import tf_slim as slim
 
 from object_detection.meta_architectures import ssd_meta_arch
 from object_detection.models import feature_map_generators
@@ -27,8 +27,6 @@ from object_detection.utils import shape_utils
 from nets.mobilenet import mobilenet
 from nets.mobilenet import mobilenet_v2
 
-slim = contrib_slim
-
 
 class SSDMobileNetV2FeatureExtractor(ssd_meta_arch.SSDFeatureExtractor):
   """SSD Feature Extractor using MobilenetV2 features."""

research/object_detection/models/ssd_mobilenet_v2_feature_extractor_test.py

@@ -17,15 +17,12 @@
 from absl.testing import parameterized
 
 import numpy as np
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
 
 from object_detection.models import ssd_feature_extractor_test
 from object_detection.models import ssd_mobilenet_v2_feature_extractor
 from object_detection.models import ssd_mobilenet_v2_keras_feature_extractor
 
-slim = contrib_slim
-
 
 @parameterized.parameters(
     {'use_keras': False},

research/object_detection/models/ssd_mobilenet_v2_fpn_feature_extractor.py

@@ -23,8 +23,8 @@ from __future__ import print_function
 import copy
 import functools
 from six.moves import range
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
+import tf_slim as slim
 
 from object_detection.meta_architectures import ssd_meta_arch
 from object_detection.models import feature_map_generators
@@ -34,8 +34,6 @@ from object_detection.utils import shape_utils
 from nets.mobilenet import mobilenet
 from nets.mobilenet import mobilenet_v2
 
-slim = contrib_slim
-
 
 # A modified config of mobilenet v2 that makes it more detection friendly.
 def _create_modified_mobilenet_config():

research/object_detection/models/ssd_mobilenet_v2_fpn_feature_extractor_test.py

@@ -20,15 +20,12 @@ Keras-based Mobilenet V2 FPN feature extractors in SSD.
 """
 from absl.testing import parameterized
 import numpy as np
-import tensorflow as tf
-from tensorflow.contrib import slim as contrib_slim
+import tensorflow.compat.v1 as tf
 
 from object_detection.models import ssd_feature_extractor_test
 from object_detection.models import ssd_mobilenet_v2_fpn_feature_extractor
 from object_detection.models import ssd_mobilenet_v2_fpn_keras_feature_extractor
 
-slim = contrib_slim
-
 
 @parameterized.parameters(
     {

research/object_detection/models/ssd_mobilenet_v2_fpn_keras_feature_extractor.py

@@ -21,7 +21,7 @@ from __future__ import division
 from __future__ import print_function
 
 from six.moves import range
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from object_detection.meta_architectures import ssd_meta_arch
 from object_detection.models import feature_map_generators