Python nets.resnet_utils.stack_blocks_dense() Examples

The following are 30 code examples of nets.resnet_utils.stack_blocks_dense(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module nets.resnet_utils , or try the search function .
Example #1
Source File: resnet_v2_test.py    From DOTA_models with Apache License 2.0 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #2
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From Gun-Detector with Apache License 2.0 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm],
                            is_training=self._train_batch_norm):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features
Example #3
Source File: resnet_v1_test.py    From ctw-baseline with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #4
Source File: resnet_v2_test.py    From ctw-baseline with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #5
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From vehicle_counting_tensorflow with MIT License 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm],
                            is_training=self._train_batch_norm):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features
Example #6
Source File: resnet_v2_test.py    From CBAM-tensorflow-slim with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #7
Source File: resnet_v2_test.py    From R3Det_Tensorflow with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #8
Source File: resnet_v1_test.py    From R3Det_Tensorflow with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #9
Source File: resnet_v1_test.py    From RetinaNet_Tensorflow_Rotation with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #10
Source File: resnet_v2_test.py    From CVTron with Apache License 2.0 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #11
Source File: resnet_v1_test.py    From CVTron with Apache License 2.0 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #12
Source File: resnet_v2_test.py    From RetinaNet_Tensorflow_Rotation with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #13
Source File: resnet_v2_test.py    From morph-net with Apache License 2.0 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #14
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From tensorflow with BSD 2-Clause "Simplified" License 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm], is_training=False):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features
Example #15
Source File: resnet_v1_test.py    From morph-net with Apache License 2.0 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #16
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From Person-Detection-and-Tracking with MIT License 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm],
                            is_training=self._train_batch_norm):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features
Example #17
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From cartoonify with MIT License 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm],
                            is_training=self._train_batch_norm):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features
Example #18
Source File: resnet_v2_test.py    From edafa with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #19
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From Hands-On-Machine-Learning-with-OpenCV-4 with MIT License 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm], is_training=False):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features
Example #20
Source File: resnet_v1_test.py    From edafa with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #21
Source File: resnet_v2_test.py    From garbage-object-detection-tensorflow with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #22
Source File: resnet_v1_test.py    From Hands-On-Machine-Learning-with-OpenCV-4 with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #23
Source File: resnet_v1_test.py    From garbage-object-detection-tensorflow with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #24
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From garbage-object-detection-tensorflow with MIT License 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm], is_training=False):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features
Example #25
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From HereIsWally with MIT License 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm], is_training=False):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features
Example #26
Source File: resnet_v2_test.py    From yolo_v2 with Apache License 2.0 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #27
Source File: resnet_v2_test.py    From Hands-On-Machine-Learning-with-OpenCV-4 with MIT License 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #28
Source File: resnet_v1_test.py    From yolo_v2 with Apache License 2.0 5 votes vote down vote up 
def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):
    """A plain ResNet without extra layers before or after the ResNet blocks."""
    with tf.variable_scope(scope, values=[inputs]):
      with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):
        net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)
        end_points = slim.utils.convert_collection_to_dict('end_points')
        return net, end_points
Example #29
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From Traffic-Rule-Violation-Detection-System with MIT License 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm],
                            is_training=self._train_batch_norm):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features
Example #30
Source File: faster_rcnn_resnet_v1_feature_extractor.py    From yolo_v2 with Apache License 2.0 5 votes vote down vote up 
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name (unused).

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(self._architecture, reuse=self._reuse_weights):
      with slim.arg_scope(
          resnet_utils.resnet_arg_scope(
              batch_norm_epsilon=1e-5,
              batch_norm_scale=True,
              weight_decay=self._weight_decay)):
        with slim.arg_scope([slim.batch_norm],
                            is_training=self._train_batch_norm):
          blocks = [
              resnet_utils.Block('block4', resnet_v1.bottleneck, [{
                  'depth': 2048,
                  'depth_bottleneck': 512,
                  'stride': 1
              }] * 3)
          ]
          proposal_classifier_features = resnet_utils.stack_blocks_dense(
              proposal_feature_maps, blocks)
    return proposal_classifier_features