exporter.py 17.2 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
# 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.
# ==============================================================================

"""Functions to export object detection inference graph."""
import logging
import os
Vivek Rathod's avatar
Vivek Rathod committed
19
import tempfile
20
import tensorflow as tf
21
from google.protobuf import text_format
22
from tensorflow.core.protobuf import saver_pb2
23
24
25
26
from tensorflow.python import pywrap_tensorflow
from tensorflow.python.client import session
from tensorflow.python.framework import graph_util
from tensorflow.python.platform import gfile
27
from tensorflow.python.saved_model import signature_constants
28
29
30
31
32
33
34
35
from tensorflow.python.training import saver as saver_lib
from object_detection.builders import model_builder
from object_detection.core import standard_fields as fields
from object_detection.data_decoders import tf_example_decoder

slim = tf.contrib.slim


36
37
# TODO: Replace with freeze_graph.freeze_graph_with_def_protos when
# newer version of Tensorflow becomes more common.
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
def freeze_graph_with_def_protos(
    input_graph_def,
    input_saver_def,
    input_checkpoint,
    output_node_names,
    restore_op_name,
    filename_tensor_name,
    clear_devices,
    initializer_nodes,
    variable_names_blacklist=''):
  """Converts all variables in a graph and checkpoint into constants."""
  del restore_op_name, filename_tensor_name  # Unused by updated loading code.

  # 'input_checkpoint' may be a prefix if we're using Saver V2 format
  if not saver_lib.checkpoint_exists(input_checkpoint):
53
54
    raise ValueError(
        'Input checkpoint "' + input_checkpoint + '" does not exist!')
55
56

  if not output_node_names:
57
58
    raise ValueError(
        'You must supply the name of a node to --output_node_names.')
59
60
61
62
63
64
65

  # Remove all the explicit device specifications for this node. This helps to
  # make the graph more portable.
  if clear_devices:
    for node in input_graph_def.node:
      node.device = ''

66
67
  with tf.Graph().as_default():
    tf.import_graph_def(input_graph_def, name='')
68
    config = tf.ConfigProto(graph_options=tf.GraphOptions())
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
    with session.Session(config=config) as sess:
      if input_saver_def:
        saver = saver_lib.Saver(saver_def=input_saver_def)
        saver.restore(sess, input_checkpoint)
      else:
        var_list = {}
        reader = pywrap_tensorflow.NewCheckpointReader(input_checkpoint)
        var_to_shape_map = reader.get_variable_to_shape_map()
        for key in var_to_shape_map:
          try:
            tensor = sess.graph.get_tensor_by_name(key + ':0')
          except KeyError:
            # This tensor doesn't exist in the graph (for example it's
            # 'global_step' or a similar housekeeping element) so skip it.
            continue
          var_list[key] = tensor
        saver = saver_lib.Saver(var_list=var_list)
        saver.restore(sess, input_checkpoint)
        if initializer_nodes:
          sess.run(initializer_nodes)

      variable_names_blacklist = (variable_names_blacklist.split(',') if
                                  variable_names_blacklist else None)
      output_graph_def = graph_util.convert_variables_to_constants(
          sess,
          input_graph_def,
          output_node_names.split(','),
          variable_names_blacklist=variable_names_blacklist)
97

98
99
100
  return output_graph_def


Vivek Rathod's avatar
Vivek Rathod committed
101
102
103
104
def replace_variable_values_with_moving_averages(graph,
                                                 current_checkpoint_file,
                                                 new_checkpoint_file):
  """Replaces variable values in the checkpoint with their moving averages.
105

Vivek Rathod's avatar
Vivek Rathod committed
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
  If the current checkpoint has shadow variables maintaining moving averages of
  the variables defined in the graph, this function generates a new checkpoint
  where the variables contain the values of their moving averages.

  Args:
    graph: a tf.Graph object.
    current_checkpoint_file: a checkpoint containing both original variables and
      their moving averages.
    new_checkpoint_file: file path to write a new checkpoint.
  """
  with graph.as_default():
    variable_averages = tf.train.ExponentialMovingAverage(0.0)
    ema_variables_to_restore = variable_averages.variables_to_restore()
    with tf.Session() as sess:
      read_saver = tf.train.Saver(ema_variables_to_restore)
      read_saver.restore(sess, current_checkpoint_file)
      write_saver = tf.train.Saver()
      write_saver.save(sess, new_checkpoint_file)


def _image_tensor_input_placeholder(input_shape=None):
  """Returns input placeholder and a 4-D uint8 image tensor."""
  if input_shape is None:
    input_shape = (None, None, None, 3)
  input_tensor = tf.placeholder(
      dtype=tf.uint8, shape=input_shape, name='image_tensor')
Derek Chow's avatar
Derek Chow committed
132
  return input_tensor, input_tensor
133

134

135
def _tf_example_input_placeholder():
Derek Chow's avatar
Derek Chow committed
136
137
138
  """Returns input that accepts a batch of strings with tf examples.

  Returns:
Vivek Rathod's avatar
Vivek Rathod committed
139
    a tuple of input placeholder and the output decoded images.
Derek Chow's avatar
Derek Chow committed
140
  """
141
142
143
144
145
146
147
  batch_tf_example_placeholder = tf.placeholder(
      tf.string, shape=[None], name='tf_example')
  def decode(tf_example_string_tensor):
    tensor_dict = tf_example_decoder.TfExampleDecoder().decode(
        tf_example_string_tensor)
    image_tensor = tensor_dict[fields.InputDataFields.image]
    return image_tensor
Derek Chow's avatar
Derek Chow committed
148
149
150
151
152
153
  return (batch_tf_example_placeholder,
          tf.map_fn(decode,
                    elems=batch_tf_example_placeholder,
                    dtype=tf.uint8,
                    parallel_iterations=32,
                    back_prop=False))
154
155


156
def _encoded_image_string_tensor_input_placeholder():
Derek Chow's avatar
Derek Chow committed
157
158
159
  """Returns input that accepts a batch of PNG or JPEG strings.

  Returns:
Vivek Rathod's avatar
Vivek Rathod committed
160
    a tuple of input placeholder and the output decoded images.
Derek Chow's avatar
Derek Chow committed
161
  """
162
163
164
165
166
167
168
169
170
  batch_image_str_placeholder = tf.placeholder(
      dtype=tf.string,
      shape=[None],
      name='encoded_image_string_tensor')
  def decode(encoded_image_string_tensor):
    image_tensor = tf.image.decode_image(encoded_image_string_tensor,
                                         channels=3)
    image_tensor.set_shape((None, None, 3))
    return image_tensor
Derek Chow's avatar
Derek Chow committed
171
172
173
174
175
176
177
  return (batch_image_str_placeholder,
          tf.map_fn(
              decode,
              elems=batch_image_str_placeholder,
              dtype=tf.uint8,
              parallel_iterations=32,
              back_prop=False))
178
179


180
input_placeholder_fn_map = {
181
182
183
    'image_tensor': _image_tensor_input_placeholder,
    'encoded_image_string_tensor':
    _encoded_image_string_tensor_input_placeholder,
184
185
186
187
    'tf_example': _tf_example_input_placeholder,
}


188
189
def _add_output_tensor_nodes(postprocessed_tensors,
                             output_collection_name='inference_op'):
190
191
192
193
194
195
196
197
198
199
  """Adds output nodes for detection boxes and scores.

  Adds the following nodes for output tensors -
    * num_detections: float32 tensor of shape [batch_size].
    * detection_boxes: float32 tensor of shape [batch_size, num_boxes, 4]
      containing detected boxes.
    * detection_scores: float32 tensor of shape [batch_size, num_boxes]
      containing scores for the detected boxes.
    * detection_classes: float32 tensor of shape [batch_size, num_boxes]
      containing class predictions for the detected boxes.
200
201
202
    * detection_masks: (Optional) float32 tensor of shape
      [batch_size, num_boxes, mask_height, mask_width] containing masks for each
      detection box.
203
204
205
206
207
208

  Args:
    postprocessed_tensors: a dictionary containing the following fields
      'detection_boxes': [batch, max_detections, 4]
      'detection_scores': [batch, max_detections]
      'detection_classes': [batch, max_detections]
209
210
      'detection_masks': [batch, max_detections, mask_height, mask_width]
        (optional).
211
      'num_detections': [batch]
212
    output_collection_name: Name of collection to add output tensors to.
213
214
215

  Returns:
    A tensor dict containing the added output tensor nodes.
216
  """
217
  detection_fields = fields.DetectionResultFields
218
  label_id_offset = 1
219
220
221
222
223
224
  boxes = postprocessed_tensors.get(detection_fields.detection_boxes)
  scores = postprocessed_tensors.get(detection_fields.detection_scores)
  classes = postprocessed_tensors.get(
      detection_fields.detection_classes) + label_id_offset
  masks = postprocessed_tensors.get(detection_fields.detection_masks)
  num_detections = postprocessed_tensors.get(detection_fields.num_detections)
225
  outputs = {}
226
227
228
229
230
231
232
233
  outputs[detection_fields.detection_boxes] = tf.identity(
      boxes, name=detection_fields.detection_boxes)
  outputs[detection_fields.detection_scores] = tf.identity(
      scores, name=detection_fields.detection_scores)
  outputs[detection_fields.detection_classes] = tf.identity(
      classes, name=detection_fields.detection_classes)
  outputs[detection_fields.num_detections] = tf.identity(
      num_detections, name=detection_fields.num_detections)
234
  if masks is not None:
235
236
    outputs[detection_fields.detection_masks] = tf.identity(
        masks, name=detection_fields.detection_masks)
237
238
239
  for output_key in outputs:
    tf.add_to_collection(output_collection_name, outputs[output_key])
  if masks is not None:
240
241
    tf.add_to_collection(output_collection_name,
                         outputs[detection_fields.detection_masks])
242
  return outputs
243
244


245
246
def _write_frozen_graph(frozen_graph_path, frozen_graph_def):
  """Writes frozen graph to disk.
247
248

  Args:
249
250
    frozen_graph_path: Path to write inference graph.
    frozen_graph_def: tf.GraphDef holding frozen graph.
251
  """
252
253
254
255
256
257
258
259
260
  with gfile.GFile(frozen_graph_path, 'wb') as f:
    f.write(frozen_graph_def.SerializeToString())
  logging.info('%d ops in the final graph.', len(frozen_graph_def.node))


def _write_saved_model(saved_model_path,
                       frozen_graph_def,
                       inputs,
                       outputs):
261
262
263
264
265
266
267
268
269
  """Writes SavedModel to disk.

  If checkpoint_path is not None bakes the weights into the graph thereby
  eliminating the need of checkpoint files during inference. If the model
  was trained with moving averages, setting use_moving_averages to true
  restores the moving averages, otherwise the original set of variables
  is restored.

  Args:
270
271
    saved_model_path: Path to write SavedModel.
    frozen_graph_def: tf.GraphDef holding frozen graph.
272
273
274
275
276
277
    inputs: The input image tensor to use for detection.
    outputs: A tensor dictionary containing the outputs of a DetectionModel.
  """
  with tf.Graph().as_default():
    with session.Session() as sess:

278
      tf.import_graph_def(frozen_graph_def, name='')
279

280
      builder = tf.saved_model.builder.SavedModelBuilder(saved_model_path)
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296

      tensor_info_inputs = {
          'inputs': tf.saved_model.utils.build_tensor_info(inputs)}
      tensor_info_outputs = {}
      for k, v in outputs.items():
        tensor_info_outputs[k] = tf.saved_model.utils.build_tensor_info(v)

      detection_signature = (
          tf.saved_model.signature_def_utils.build_signature_def(
              inputs=tensor_info_inputs,
              outputs=tensor_info_outputs,
              method_name=signature_constants.PREDICT_METHOD_NAME))

      builder.add_meta_graph_and_variables(
          sess, [tf.saved_model.tag_constants.SERVING],
          signature_def_map={
Derek Chow's avatar
Derek Chow committed
297
              signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:
298
299
300
301
302
303
                  detection_signature,
          },
      )
      builder.save()


304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
def _write_graph_and_checkpoint(inference_graph_def,
                                model_path,
                                input_saver_def,
                                trained_checkpoint_prefix):
  for node in inference_graph_def.node:
    node.device = ''
  with tf.Graph().as_default():
    tf.import_graph_def(inference_graph_def, name='')
    with session.Session() as sess:
      saver = saver_lib.Saver(saver_def=input_saver_def,
                              save_relative_paths=True)
      saver.restore(sess, trained_checkpoint_prefix)
      saver.save(sess, model_path)


319
320
321
def _export_inference_graph(input_type,
                            detection_model,
                            use_moving_averages,
322
323
                            trained_checkpoint_prefix,
                            output_directory,
Vivek Rathod's avatar
Vivek Rathod committed
324
325
                            additional_output_tensor_names=None,
                            input_shape=None,
326
327
                            output_collection_name='inference_op',
                            graph_hook_fn=None):
328
  """Export helper."""
329
330
331
332
333
334
  tf.gfile.MakeDirs(output_directory)
  frozen_graph_path = os.path.join(output_directory,
                                   'frozen_inference_graph.pb')
  saved_model_path = os.path.join(output_directory, 'saved_model')
  model_path = os.path.join(output_directory, 'model.ckpt')

335
336
  if input_type not in input_placeholder_fn_map:
    raise ValueError('Unknown input type: {}'.format(input_type))
Vivek Rathod's avatar
Vivek Rathod committed
337
338
339
340
341
342
343
344
  placeholder_args = {}
  if input_shape is not None:
    if input_type != 'image_tensor':
      raise ValueError('Can only specify input shape for `image_tensor` '
                       'inputs.')
    placeholder_args['input_shape'] = input_shape
  placeholder_tensor, input_tensors = input_placeholder_fn_map[input_type](
      **placeholder_args)
Derek Chow's avatar
Derek Chow committed
345
  inputs = tf.to_float(input_tensors)
346
347
348
349
350
  preprocessed_inputs, true_image_shapes = detection_model.preprocess(inputs)
  output_tensors = detection_model.predict(
      preprocessed_inputs, true_image_shapes)
  postprocessed_tensors = detection_model.postprocess(
      output_tensors, true_image_shapes)
351
352
  outputs = _add_output_tensor_nodes(postprocessed_tensors,
                                     output_collection_name)
Vivek Rathod's avatar
Vivek Rathod committed
353
354
  # Add global step to the graph.
  slim.get_or_create_global_step()
355

356
357
  if graph_hook_fn: graph_hook_fn()

358
  saver_kwargs = {}
359
  if use_moving_averages:
360
361
362
363
364
365
    # This check is to be compatible with both version of SaverDef.
    if os.path.isfile(trained_checkpoint_prefix):
      saver_kwargs['write_version'] = saver_pb2.SaverDef.V1
      temp_checkpoint_prefix = tempfile.NamedTemporaryFile().name
    else:
      temp_checkpoint_prefix = tempfile.mkdtemp()
Vivek Rathod's avatar
Vivek Rathod committed
366
367
    replace_variable_values_with_moving_averages(
        tf.get_default_graph(), trained_checkpoint_prefix,
368
369
        temp_checkpoint_prefix)
    checkpoint_to_use = temp_checkpoint_prefix
370
  else:
Vivek Rathod's avatar
Vivek Rathod committed
371
372
    checkpoint_to_use = trained_checkpoint_prefix

373
  saver = tf.train.Saver(**saver_kwargs)
374
375
376
377
378
379
  input_saver_def = saver.as_saver_def()

  _write_graph_and_checkpoint(
      inference_graph_def=tf.get_default_graph().as_graph_def(),
      model_path=model_path,
      input_saver_def=input_saver_def,
Vivek Rathod's avatar
Vivek Rathod committed
380
381
382
383
384
385
      trained_checkpoint_prefix=checkpoint_to_use)

  if additional_output_tensor_names is not None:
    output_node_names = ','.join(outputs.keys()+additional_output_tensor_names)
  else:
    output_node_names = ','.join(outputs.keys())
386
387
388
389

  frozen_graph_def = freeze_graph_with_def_protos(
      input_graph_def=tf.get_default_graph().as_graph_def(),
      input_saver_def=input_saver_def,
Vivek Rathod's avatar
Vivek Rathod committed
390
391
      input_checkpoint=checkpoint_to_use,
      output_node_names=output_node_names,
392
393
394
395
396
      restore_op_name='save/restore_all',
      filename_tensor_name='save/Const:0',
      clear_devices=True,
      initializer_nodes='')
  _write_frozen_graph(frozen_graph_path, frozen_graph_def)
Vivek Rathod's avatar
Vivek Rathod committed
397
398
  _write_saved_model(saved_model_path, frozen_graph_def,
                     placeholder_tensor, outputs)
399
400


401
402
403
404
def export_inference_graph(input_type,
                           pipeline_config,
                           trained_checkpoint_prefix,
                           output_directory,
Vivek Rathod's avatar
Vivek Rathod committed
405
406
407
                           input_shape=None,
                           output_collection_name='inference_op',
                           additional_output_tensor_names=None):
408
409
410
411
412
413
  """Exports inference graph for the model specified in the pipeline config.

  Args:
    input_type: Type of input for the graph. Can be one of [`image_tensor`,
      `tf_example`].
    pipeline_config: pipeline_pb2.TrainAndEvalPipelineConfig proto.
414
415
    trained_checkpoint_prefix: Path to the trained checkpoint file.
    output_directory: Path to write outputs.
Vivek Rathod's avatar
Vivek Rathod committed
416
417
    input_shape: Sets a fixed shape for an `image_tensor` input. If not
      specified, will default to [None, None, None, 3].
418
419
    output_collection_name: Name of collection to add output tensors to.
      If None, does not add output tensors to a collection.
Vivek Rathod's avatar
Vivek Rathod committed
420
    additional_output_tensor_names: list of additional output
421
      tensors to include in the frozen graph.
422
423
424
425
426
  """
  detection_model = model_builder.build(pipeline_config.model,
                                        is_training=False)
  _export_inference_graph(input_type, detection_model,
                          pipeline_config.eval_config.use_moving_averages,
Vivek Rathod's avatar
Vivek Rathod committed
427
428
                          trained_checkpoint_prefix,
                          output_directory, additional_output_tensor_names,
429
430
431
432
433
434
435
                          input_shape, output_collection_name,
                          graph_hook_fn=None)
  pipeline_config.eval_config.use_moving_averages = False
  config_text = text_format.MessageToString(pipeline_config)
  with tf.gfile.Open(
      os.path.join(output_directory, 'pipeline.config'), 'wb') as f:
    f.write(config_text)