inputs_test.py 69.5 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
# 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 object_detection.tflearn.inputs."""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

21
import functools
22
import os
23
import unittest
24
from absl import logging
pkulzc's avatar
pkulzc committed
25
from absl.testing import parameterized
26
import numpy as np
27
import six
28
import tensorflow.compat.v1 as tf
29
30

from object_detection import inputs
31
from object_detection.core import preprocessor
32
33
from object_detection.core import standard_fields as fields
from object_detection.utils import config_util
pkulzc's avatar
pkulzc committed
34
from object_detection.utils import test_case
35
36
37
38
39
40
41
from object_detection.utils import test_utils
from object_detection.utils import tf_version

if six.PY2:
  import mock  # pylint: disable=g-import-not-at-top
else:
  from unittest import mock  # pylint: disable=g-import-not-at-top, g-importing-member
42
43
44
45
46
47

FLAGS = tf.flags.FLAGS


def _get_configs_for_model(model_name):
  """Returns configurations for model."""
Zhichao Lu's avatar
Zhichao Lu committed
48
49
50
51
52
53
  fname = os.path.join(tf.resource_loader.get_data_files_path(),
                       'samples/configs/' + model_name + '.config')
  label_map_path = os.path.join(tf.resource_loader.get_data_files_path(),
                                'data/pet_label_map.pbtxt')
  data_path = os.path.join(tf.resource_loader.get_data_files_path(),
                           'test_data/pets_examples.record')
54
  configs = config_util.get_configs_from_pipeline_file(fname)
55
56
57
58
59
  override_dict = {
      'train_input_path': data_path,
      'eval_input_path': data_path,
      'label_map_path': label_map_path
  }
60
  return config_util.merge_external_params_with_configs(
61
      configs, kwargs_dict=override_dict)
62
63


64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
def _get_configs_for_model_sequence_example(model_name):
  """Returns configurations for model."""
  fname = os.path.join(tf.resource_loader.get_data_files_path(),
                       'test_data/' + model_name + '.config')
  label_map_path = os.path.join(tf.resource_loader.get_data_files_path(),
                                'data/snapshot_serengeti_label_map.pbtxt')
  data_path = os.path.join(
      tf.resource_loader.get_data_files_path(),
      'test_data/snapshot_serengeti_sequence_examples.record')
  configs = config_util.get_configs_from_pipeline_file(fname)
  override_dict = {
      'train_input_path': data_path,
      'eval_input_path': data_path,
      'label_map_path': label_map_path
  }
  return config_util.merge_external_params_with_configs(
      configs, kwargs_dict=override_dict)


83
84
85
86
87
88
89
90
91
def _make_initializable_iterator(dataset):
  """Creates an iterator, and initializes tables.

  Args:
    dataset: A `tf.data.Dataset` object.

  Returns:
    A `tf.data.Iterator`.
  """
92
  iterator = tf.data.make_initializable_iterator(dataset)
93
94
95
96
  tf.add_to_collection(tf.GraphKeys.TABLE_INITIALIZERS, iterator.initializer)
  return iterator


97
98
@unittest.skipIf(tf_version.is_tf2(), 'Skipping TF1.X only tests under TF2.X.')
class InputFnTest(test_case.TestCase, parameterized.TestCase):
99
100
101
102

  def test_faster_rcnn_resnet50_train_input(self):
    """Tests the training input function for FasterRcnnResnet50."""
    configs = _get_configs_for_model('faster_rcnn_resnet50_pets')
103
104
    model_config = configs['model']
    model_config.faster_rcnn.num_classes = 37
105
    train_input_fn = inputs.create_train_input_fn(
106
        configs['train_config'], configs['train_input_config'], model_config)
107
    features, labels = _make_initializable_iterator(train_input_fn()).get_next()
108

109
    self.assertAllEqual([1, None, None, 3],
110
111
                        features[fields.InputDataFields.image].shape.as_list())
    self.assertEqual(tf.float32, features[fields.InputDataFields.image].dtype)
112
    self.assertAllEqual([1],
113
114
115
                        features[inputs.HASH_KEY].shape.as_list())
    self.assertEqual(tf.int32, features[inputs.HASH_KEY].dtype)
    self.assertAllEqual(
116
        [1, 100, 4],
117
118
119
120
        labels[fields.InputDataFields.groundtruth_boxes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_boxes].dtype)
    self.assertAllEqual(
121
        [1, 100, model_config.faster_rcnn.num_classes],
122
123
124
        labels[fields.InputDataFields.groundtruth_classes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_classes].dtype)
125
126
127
128
129
    self.assertAllEqual(
        [1, 100],
        labels[fields.InputDataFields.groundtruth_weights].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_weights].dtype)
130
131
132
133
134
135
    self.assertAllEqual(
        [1, 100, model_config.faster_rcnn.num_classes],
        labels[fields.InputDataFields.groundtruth_confidences].shape.as_list())
    self.assertEqual(
        tf.float32,
        labels[fields.InputDataFields.groundtruth_confidences].dtype)
136

137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
  def test_faster_rcnn_resnet50_train_input_with_additional_channels(self):
    """Tests the training input function for FasterRcnnResnet50."""
    configs = _get_configs_for_model('faster_rcnn_resnet50_pets')
    model_config = configs['model']
    configs['train_input_config'].num_additional_channels = 2
    configs['train_config'].retain_original_images = True
    model_config.faster_rcnn.num_classes = 37
    train_input_fn = inputs.create_train_input_fn(
        configs['train_config'], configs['train_input_config'], model_config)
    features, labels = _make_initializable_iterator(train_input_fn()).get_next()

    self.assertAllEqual([1, None, None, 5],
                        features[fields.InputDataFields.image].shape.as_list())
    self.assertAllEqual(
        [1, None, None, 3],
        features[fields.InputDataFields.original_image].shape.as_list())
    self.assertEqual(tf.float32, features[fields.InputDataFields.image].dtype)
    self.assertAllEqual([1],
                        features[inputs.HASH_KEY].shape.as_list())
    self.assertEqual(tf.int32, features[inputs.HASH_KEY].dtype)
    self.assertAllEqual(
        [1, 100, 4],
        labels[fields.InputDataFields.groundtruth_boxes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_boxes].dtype)
    self.assertAllEqual(
        [1, 100, model_config.faster_rcnn.num_classes],
        labels[fields.InputDataFields.groundtruth_classes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_classes].dtype)
    self.assertAllEqual(
        [1, 100],
        labels[fields.InputDataFields.groundtruth_weights].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_weights].dtype)
    self.assertAllEqual(
        [1, 100, model_config.faster_rcnn.num_classes],
        labels[fields.InputDataFields.groundtruth_confidences].shape.as_list())
    self.assertEqual(
        tf.float32,
        labels[fields.InputDataFields.groundtruth_confidences].dtype)

pkulzc's avatar
pkulzc committed
179
180
181
182
183
  @parameterized.parameters(
      {'eval_batch_size': 1},
      {'eval_batch_size': 8}
  )
  def test_faster_rcnn_resnet50_eval_input(self, eval_batch_size=1):
184
185
    """Tests the eval input function for FasterRcnnResnet50."""
    configs = _get_configs_for_model('faster_rcnn_resnet50_pets')
186
187
    model_config = configs['model']
    model_config.faster_rcnn.num_classes = 37
pkulzc's avatar
pkulzc committed
188
189
    eval_config = configs['eval_config']
    eval_config.batch_size = eval_batch_size
190
    eval_input_fn = inputs.create_eval_input_fn(
pkulzc's avatar
pkulzc committed
191
        eval_config, configs['eval_input_configs'][0], model_config)
192
    features, labels = _make_initializable_iterator(eval_input_fn()).get_next()
pkulzc's avatar
pkulzc committed
193
    self.assertAllEqual([eval_batch_size, None, None, 3],
194
195
196
                        features[fields.InputDataFields.image].shape.as_list())
    self.assertEqual(tf.float32, features[fields.InputDataFields.image].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
197
        [eval_batch_size, None, None, 3],
198
199
200
        features[fields.InputDataFields.original_image].shape.as_list())
    self.assertEqual(tf.uint8,
                     features[fields.InputDataFields.original_image].dtype)
pkulzc's avatar
pkulzc committed
201
202
    self.assertAllEqual([eval_batch_size],
                        features[inputs.HASH_KEY].shape.as_list())
203
204
    self.assertEqual(tf.int32, features[inputs.HASH_KEY].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
205
        [eval_batch_size, 100, 4],
206
207
208
209
        labels[fields.InputDataFields.groundtruth_boxes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_boxes].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
210
        [eval_batch_size, 100, model_config.faster_rcnn.num_classes],
211
212
213
        labels[fields.InputDataFields.groundtruth_classes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_classes].dtype)
214
    self.assertAllEqual(
215
216
        [eval_batch_size, 100],
        labels[fields.InputDataFields.groundtruth_weights].shape.as_list())
217
218
    self.assertEqual(
        tf.float32,
219
        labels[fields.InputDataFields.groundtruth_weights].dtype)
220
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
221
        [eval_batch_size, 100],
222
223
224
225
        labels[fields.InputDataFields.groundtruth_area].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_area].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
226
        [eval_batch_size, 100],
227
228
229
230
        labels[fields.InputDataFields.groundtruth_is_crowd].shape.as_list())
    self.assertEqual(
        tf.bool, labels[fields.InputDataFields.groundtruth_is_crowd].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
231
        [eval_batch_size, 100],
232
233
234
        labels[fields.InputDataFields.groundtruth_difficult].shape.as_list())
    self.assertEqual(
        tf.int32, labels[fields.InputDataFields.groundtruth_difficult].dtype)
235

236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
  def test_context_rcnn_resnet50_train_input_with_sequence_example(
      self, train_batch_size=8):
    """Tests the training input function for FasterRcnnResnet50."""
    configs = _get_configs_for_model_sequence_example(
        'context_rcnn_camera_trap')
    model_config = configs['model']
    train_config = configs['train_config']
    train_config.batch_size = train_batch_size
    train_input_fn = inputs.create_train_input_fn(
        train_config, configs['train_input_config'], model_config)
    features, labels = _make_initializable_iterator(train_input_fn()).get_next()

    self.assertAllEqual([train_batch_size, 640, 640, 3],
                        features[fields.InputDataFields.image].shape.as_list())
    self.assertEqual(tf.float32, features[fields.InputDataFields.image].dtype)
    self.assertAllEqual([train_batch_size],
                        features[inputs.HASH_KEY].shape.as_list())
    self.assertEqual(tf.int32, features[inputs.HASH_KEY].dtype)
    self.assertAllEqual(
        [train_batch_size, 100, 4],
        labels[fields.InputDataFields.groundtruth_boxes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_boxes].dtype)
    self.assertAllEqual(
        [train_batch_size, 100, model_config.faster_rcnn.num_classes],
        labels[fields.InputDataFields.groundtruth_classes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_classes].dtype)
    self.assertAllEqual(
        [train_batch_size, 100],
        labels[fields.InputDataFields.groundtruth_weights].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_weights].dtype)
    self.assertAllEqual(
        [train_batch_size, 100, model_config.faster_rcnn.num_classes],
        labels[fields.InputDataFields.groundtruth_confidences].shape.as_list())
    self.assertEqual(
        tf.float32,
        labels[fields.InputDataFields.groundtruth_confidences].dtype)

  def test_context_rcnn_resnet50_eval_input_with_sequence_example(
      self, eval_batch_size=8):
    """Tests the eval input function for FasterRcnnResnet50."""
    configs = _get_configs_for_model_sequence_example(
        'context_rcnn_camera_trap')
    model_config = configs['model']
    eval_config = configs['eval_config']
    eval_config.batch_size = eval_batch_size
    eval_input_fn = inputs.create_eval_input_fn(
        eval_config, configs['eval_input_configs'][0], model_config)
    features, labels = _make_initializable_iterator(eval_input_fn()).get_next()
    self.assertAllEqual([eval_batch_size, 640, 640, 3],
                        features[fields.InputDataFields.image].shape.as_list())
    self.assertEqual(tf.float32, features[fields.InputDataFields.image].dtype)
    self.assertAllEqual(
        [eval_batch_size, 640, 640, 3],
        features[fields.InputDataFields.original_image].shape.as_list())
    self.assertEqual(tf.uint8,
                     features[fields.InputDataFields.original_image].dtype)
    self.assertAllEqual([eval_batch_size],
                        features[inputs.HASH_KEY].shape.as_list())
    self.assertEqual(tf.int32, features[inputs.HASH_KEY].dtype)
    self.assertAllEqual(
        [eval_batch_size, 100, 4],
        labels[fields.InputDataFields.groundtruth_boxes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_boxes].dtype)
    self.assertAllEqual(
        [eval_batch_size, 100, model_config.faster_rcnn.num_classes],
        labels[fields.InputDataFields.groundtruth_classes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_classes].dtype)
    self.assertAllEqual(
        [eval_batch_size, 100],
        labels[fields.InputDataFields.groundtruth_weights].shape.as_list())
    self.assertEqual(
        tf.float32,
        labels[fields.InputDataFields.groundtruth_weights].dtype)

315
316
317
  def test_ssd_inceptionV2_train_input(self):
    """Tests the training input function for SSDInceptionV2."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
318
319
    model_config = configs['model']
    model_config.ssd.num_classes = 37
320
321
    batch_size = configs['train_config'].batch_size
    train_input_fn = inputs.create_train_input_fn(
322
        configs['train_config'], configs['train_input_config'], model_config)
323
    features, labels = _make_initializable_iterator(train_input_fn()).get_next()
324
325
326
327
328
329
330
331
332
333
334
335
336

    self.assertAllEqual([batch_size, 300, 300, 3],
                        features[fields.InputDataFields.image].shape.as_list())
    self.assertEqual(tf.float32, features[fields.InputDataFields.image].dtype)
    self.assertAllEqual([batch_size],
                        features[inputs.HASH_KEY].shape.as_list())
    self.assertEqual(tf.int32, features[inputs.HASH_KEY].dtype)
    self.assertAllEqual(
        [batch_size],
        labels[fields.InputDataFields.num_groundtruth_boxes].shape.as_list())
    self.assertEqual(tf.int32,
                     labels[fields.InputDataFields.num_groundtruth_boxes].dtype)
    self.assertAllEqual(
337
        [batch_size, 100, 4],
338
339
340
341
        labels[fields.InputDataFields.groundtruth_boxes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_boxes].dtype)
    self.assertAllEqual(
342
        [batch_size, 100, model_config.ssd.num_classes],
343
344
345
        labels[fields.InputDataFields.groundtruth_classes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_classes].dtype)
346
    self.assertAllEqual(
347
        [batch_size, 100],
348
        labels[
349
            fields.InputDataFields.groundtruth_weights].shape.as_list())
350
351
    self.assertEqual(
        tf.float32,
352
        labels[fields.InputDataFields.groundtruth_weights].dtype)
353

pkulzc's avatar
pkulzc committed
354
355
356
357
358
  @parameterized.parameters(
      {'eval_batch_size': 1},
      {'eval_batch_size': 8}
  )
  def test_ssd_inceptionV2_eval_input(self, eval_batch_size=1):
359
360
    """Tests the eval input function for SSDInceptionV2."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
361
362
    model_config = configs['model']
    model_config.ssd.num_classes = 37
pkulzc's avatar
pkulzc committed
363
364
    eval_config = configs['eval_config']
    eval_config.batch_size = eval_batch_size
365
    eval_input_fn = inputs.create_eval_input_fn(
pkulzc's avatar
pkulzc committed
366
        eval_config, configs['eval_input_configs'][0], model_config)
367
    features, labels = _make_initializable_iterator(eval_input_fn()).get_next()
pkulzc's avatar
pkulzc committed
368
    self.assertAllEqual([eval_batch_size, 300, 300, 3],
369
370
371
                        features[fields.InputDataFields.image].shape.as_list())
    self.assertEqual(tf.float32, features[fields.InputDataFields.image].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
372
        [eval_batch_size, 300, 300, 3],
373
374
375
        features[fields.InputDataFields.original_image].shape.as_list())
    self.assertEqual(tf.uint8,
                     features[fields.InputDataFields.original_image].dtype)
pkulzc's avatar
pkulzc committed
376
377
    self.assertAllEqual([eval_batch_size],
                        features[inputs.HASH_KEY].shape.as_list())
378
379
    self.assertEqual(tf.int32, features[inputs.HASH_KEY].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
380
        [eval_batch_size, 100, 4],
381
382
383
384
        labels[fields.InputDataFields.groundtruth_boxes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_boxes].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
385
        [eval_batch_size, 100, model_config.ssd.num_classes],
386
387
388
        labels[fields.InputDataFields.groundtruth_classes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_classes].dtype)
389
    self.assertAllEqual(
390
        [eval_batch_size, 100],
391
        labels[
392
            fields.InputDataFields.groundtruth_weights].shape.as_list())
393
394
    self.assertEqual(
        tf.float32,
395
        labels[fields.InputDataFields.groundtruth_weights].dtype)
396
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
397
        [eval_batch_size, 100],
398
399
400
401
        labels[fields.InputDataFields.groundtruth_area].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_area].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
402
        [eval_batch_size, 100],
403
404
405
406
        labels[fields.InputDataFields.groundtruth_is_crowd].shape.as_list())
    self.assertEqual(
        tf.bool, labels[fields.InputDataFields.groundtruth_is_crowd].dtype)
    self.assertAllEqual(
pkulzc's avatar
pkulzc committed
407
        [eval_batch_size, 100],
408
409
410
        labels[fields.InputDataFields.groundtruth_difficult].shape.as_list())
    self.assertEqual(
        tf.int32, labels[fields.InputDataFields.groundtruth_difficult].dtype)
411

412
413
  def test_ssd_inceptionV2_eval_input_with_additional_channels(
      self, eval_batch_size=1):
414
    """Tests the eval input function for SSDInceptionV2 with additional channel.
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475

    Args:
      eval_batch_size: Batch size for eval set.
    """
    configs = _get_configs_for_model('ssd_inception_v2_pets')
    model_config = configs['model']
    model_config.ssd.num_classes = 37
    configs['eval_input_configs'][0].num_additional_channels = 1
    eval_config = configs['eval_config']
    eval_config.batch_size = eval_batch_size
    eval_config.retain_original_image_additional_channels = True
    eval_input_fn = inputs.create_eval_input_fn(
        eval_config, configs['eval_input_configs'][0], model_config)
    features, labels = _make_initializable_iterator(eval_input_fn()).get_next()
    self.assertAllEqual([eval_batch_size, 300, 300, 4],
                        features[fields.InputDataFields.image].shape.as_list())
    self.assertEqual(tf.float32, features[fields.InputDataFields.image].dtype)
    self.assertAllEqual(
        [eval_batch_size, 300, 300, 3],
        features[fields.InputDataFields.original_image].shape.as_list())
    self.assertEqual(tf.uint8,
                     features[fields.InputDataFields.original_image].dtype)
    self.assertAllEqual([eval_batch_size, 300, 300, 1], features[
        fields.InputDataFields.image_additional_channels].shape.as_list())
    self.assertEqual(
        tf.uint8,
        features[fields.InputDataFields.image_additional_channels].dtype)
    self.assertAllEqual([eval_batch_size],
                        features[inputs.HASH_KEY].shape.as_list())
    self.assertEqual(tf.int32, features[inputs.HASH_KEY].dtype)
    self.assertAllEqual(
        [eval_batch_size, 100, 4],
        labels[fields.InputDataFields.groundtruth_boxes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_boxes].dtype)
    self.assertAllEqual(
        [eval_batch_size, 100, model_config.ssd.num_classes],
        labels[fields.InputDataFields.groundtruth_classes].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_classes].dtype)
    self.assertAllEqual(
        [eval_batch_size, 100],
        labels[fields.InputDataFields.groundtruth_weights].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_weights].dtype)
    self.assertAllEqual(
        [eval_batch_size, 100],
        labels[fields.InputDataFields.groundtruth_area].shape.as_list())
    self.assertEqual(tf.float32,
                     labels[fields.InputDataFields.groundtruth_area].dtype)
    self.assertAllEqual(
        [eval_batch_size, 100],
        labels[fields.InputDataFields.groundtruth_is_crowd].shape.as_list())
    self.assertEqual(tf.bool,
                     labels[fields.InputDataFields.groundtruth_is_crowd].dtype)
    self.assertAllEqual(
        [eval_batch_size, 100],
        labels[fields.InputDataFields.groundtruth_difficult].shape.as_list())
    self.assertEqual(tf.int32,
                     labels[fields.InputDataFields.groundtruth_difficult].dtype)

476
477
  def test_predict_input(self):
    """Tests the predict input function."""
478
479
    configs = _get_configs_for_model('ssd_inception_v2_pets')
    predict_input_fn = inputs.create_predict_input_fn(
480
        model_config=configs['model'],
481
        predict_input_config=configs['eval_input_configs'][0])
482
483
    serving_input_receiver = predict_input_fn()

484
    image = serving_input_receiver.features[fields.InputDataFields.image]
485
    receiver_tensors = serving_input_receiver.receiver_tensors[
486
487
        inputs.SERVING_FED_EXAMPLE_KEY]
    self.assertEqual([1, 300, 300, 3], image.shape.as_list())
488
489
490
    self.assertEqual(tf.float32, image.dtype)
    self.assertEqual(tf.string, receiver_tensors.dtype)

491
492
493
  def test_predict_input_with_additional_channels(self):
    """Tests the predict input function with additional channels."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
494
    configs['eval_input_configs'][0].num_additional_channels = 2
495
496
    predict_input_fn = inputs.create_predict_input_fn(
        model_config=configs['model'],
497
        predict_input_config=configs['eval_input_configs'][0])
498
499
500
501
502
503
504
505
506
507
    serving_input_receiver = predict_input_fn()

    image = serving_input_receiver.features[fields.InputDataFields.image]
    receiver_tensors = serving_input_receiver.receiver_tensors[
        inputs.SERVING_FED_EXAMPLE_KEY]
    # RGB + 2 additional channels = 5 channels.
    self.assertEqual([1, 300, 300, 5], image.shape.as_list())
    self.assertEqual(tf.float32, image.dtype)
    self.assertEqual(tf.string, receiver_tensors.dtype)

508
509
510
  def test_error_with_bad_train_config(self):
    """Tests that a TypeError is raised with improper train config."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
511
    configs['model'].ssd.num_classes = 37
512
513
    train_input_fn = inputs.create_train_input_fn(
        train_config=configs['eval_config'],  # Expecting `TrainConfig`.
514
515
        train_input_config=configs['train_input_config'],
        model_config=configs['model'])
516
517
518
519
520
521
    with self.assertRaises(TypeError):
      train_input_fn()

  def test_error_with_bad_train_input_config(self):
    """Tests that a TypeError is raised with improper train input config."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
522
523
524
525
526
527
528
529
530
531
532
533
    configs['model'].ssd.num_classes = 37
    train_input_fn = inputs.create_train_input_fn(
        train_config=configs['train_config'],
        train_input_config=configs['model'],  # Expecting `InputReader`.
        model_config=configs['model'])
    with self.assertRaises(TypeError):
      train_input_fn()

  def test_error_with_bad_train_model_config(self):
    """Tests that a TypeError is raised with improper train model config."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
    configs['model'].ssd.num_classes = 37
534
535
    train_input_fn = inputs.create_train_input_fn(
        train_config=configs['train_config'],
536
537
        train_input_config=configs['train_input_config'],
        model_config=configs['train_config'])  # Expecting `DetectionModel`.
538
539
540
541
542
543
    with self.assertRaises(TypeError):
      train_input_fn()

  def test_error_with_bad_eval_config(self):
    """Tests that a TypeError is raised with improper eval config."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
544
    configs['model'].ssd.num_classes = 37
545
546
    eval_input_fn = inputs.create_eval_input_fn(
        eval_config=configs['train_config'],  # Expecting `EvalConfig`.
547
        eval_input_config=configs['eval_input_configs'][0],
548
        model_config=configs['model'])
549
550
551
552
553
554
    with self.assertRaises(TypeError):
      eval_input_fn()

  def test_error_with_bad_eval_input_config(self):
    """Tests that a TypeError is raised with improper eval input config."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
555
    configs['model'].ssd.num_classes = 37
556
557
    eval_input_fn = inputs.create_eval_input_fn(
        eval_config=configs['eval_config'],
558
559
        eval_input_config=configs['model'],  # Expecting `InputReader`.
        model_config=configs['model'])
560
561
562
    with self.assertRaises(TypeError):
      eval_input_fn()

563
564
565
566
567
568
  def test_error_with_bad_eval_model_config(self):
    """Tests that a TypeError is raised with improper eval model config."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
    configs['model'].ssd.num_classes = 37
    eval_input_fn = inputs.create_eval_input_fn(
        eval_config=configs['eval_config'],
569
        eval_input_config=configs['eval_input_configs'][0],
570
571
572
573
        model_config=configs['eval_config'])  # Expecting `DetectionModel`.
    with self.assertRaises(TypeError):
      eval_input_fn()

574
575
576
577
578
  def test_output_equal_in_replace_empty_string_with_random_number(self):
    string_placeholder = tf.placeholder(tf.string, shape=[])
    replaced_string = inputs._replace_empty_string_with_random_number(
        string_placeholder)

579
    test_string = b'hello world'
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
    feed_dict = {string_placeholder: test_string}

    with self.test_session() as sess:
      out_string = sess.run(replaced_string, feed_dict=feed_dict)

    self.assertEqual(test_string, out_string)

  def test_output_is_integer_in_replace_empty_string_with_random_number(self):

    string_placeholder = tf.placeholder(tf.string, shape=[])
    replaced_string = inputs._replace_empty_string_with_random_number(
        string_placeholder)

    empty_string = ''
    feed_dict = {string_placeholder: empty_string}
    with self.test_session() as sess:
      out_string = sess.run(replaced_string, feed_dict=feed_dict)

598
599
600
601
602
603
604
    is_integer = True
    try:
      # Test whether out_string is a string which represents an integer, the
      # casting below will throw an error if out_string is not castable to int.
      int(out_string)
    except ValueError:
      is_integer = False
605

606
    self.assertTrue(is_integer)
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633

  def test_force_no_resize(self):
    """Tests the functionality of force_no_reisze option."""
    configs = _get_configs_for_model('ssd_inception_v2_pets')
    configs['eval_config'].force_no_resize = True

    eval_input_fn = inputs.create_eval_input_fn(
        eval_config=configs['eval_config'],
        eval_input_config=configs['eval_input_configs'][0],
        model_config=configs['model']
    )
    train_input_fn = inputs.create_train_input_fn(
        train_config=configs['train_config'],
        train_input_config=configs['train_input_config'],
        model_config=configs['model']
    )

    features_train, _ = _make_initializable_iterator(
        train_input_fn()).get_next()

    features_eval, _ = _make_initializable_iterator(
        eval_input_fn()).get_next()

    images_train, images_eval = features_train['image'], features_eval['image']

    self.assertEqual([1, None, None, 3], images_eval.shape.as_list())
    self.assertEqual([24, 300, 300, 3], images_train.shape.as_list())
634

635

pkulzc's avatar
pkulzc committed
636
class DataAugmentationFnTest(test_case.TestCase):
637
638
639
640
641
642
643
644
645
646
647
648
649

  def test_apply_image_and_box_augmentation(self):
    data_augmentation_options = [
        (preprocessor.resize_image, {
            'new_height': 20,
            'new_width': 20,
            'method': tf.image.ResizeMethod.NEAREST_NEIGHBOR
        }),
        (preprocessor.scale_boxes_to_pixel_coordinates, {}),
    ]
    data_augmentation_fn = functools.partial(
        inputs.augment_input_data,
        data_augmentation_options=data_augmentation_options)
650
651
652
653
654
655
656
657
658
659
660
661
662
663
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(10, 10, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1., 1.]], np.float32))
      }
      augmented_tensor_dict = data_augmentation_fn(tensor_dict=tensor_dict)
      return (augmented_tensor_dict[fields.InputDataFields.image],
              augmented_tensor_dict[fields.InputDataFields.
                                    groundtruth_boxes])
    image, groundtruth_boxes = self.execute_cpu(graph_fn, [])
    self.assertAllEqual(image.shape, [20, 20, 3])
    self.assertAllClose(groundtruth_boxes, [[10, 10, 20, 20]])
664

665
666
667
668
669
670
671
672
673
674
675
676
  def test_apply_image_and_box_augmentation_with_scores(self):
    data_augmentation_options = [
        (preprocessor.resize_image, {
            'new_height': 20,
            'new_width': 20,
            'method': tf.image.ResizeMethod.NEAREST_NEIGHBOR
        }),
        (preprocessor.scale_boxes_to_pixel_coordinates, {}),
    ]
    data_augmentation_fn = functools.partial(
        inputs.augment_input_data,
        data_augmentation_options=data_augmentation_options)
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(10, 10, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1., 1.]], np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([1.0], np.float32)),
          fields.InputDataFields.groundtruth_weights:
              tf.constant(np.array([0.8], np.float32)),
      }
      augmented_tensor_dict = data_augmentation_fn(tensor_dict=tensor_dict)
      return (augmented_tensor_dict[fields.InputDataFields.image],
              augmented_tensor_dict[fields.InputDataFields.groundtruth_boxes],
              augmented_tensor_dict[fields.InputDataFields.groundtruth_classes],
              augmented_tensor_dict[fields.InputDataFields.groundtruth_weights])
    (image, groundtruth_boxes,
     groundtruth_classes, groundtruth_weights) = self.execute_cpu(graph_fn, [])
    self.assertAllEqual(image.shape, [20, 20, 3])
    self.assertAllClose(groundtruth_boxes, [[10, 10, 20, 20]])
    self.assertAllClose(groundtruth_classes.shape, [1.0])
    self.assertAllClose(groundtruth_weights, [0.8])
699

700
701
702
703
704
705
706
707
708
709
710
  def test_include_masks_in_data_augmentation(self):
    data_augmentation_options = [
        (preprocessor.resize_image, {
            'new_height': 20,
            'new_width': 20,
            'method': tf.image.ResizeMethod.NEAREST_NEIGHBOR
        })
    ]
    data_augmentation_fn = functools.partial(
        inputs.augment_input_data,
        data_augmentation_options=data_augmentation_options)
711
712
713
714
715
716
717
718
719
720
721
722
723
724
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(10, 10, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_instance_masks:
              tf.constant(np.zeros([2, 10, 10], np.uint8))
      }
      augmented_tensor_dict = data_augmentation_fn(tensor_dict=tensor_dict)
      return (augmented_tensor_dict[fields.InputDataFields.image],
              augmented_tensor_dict[fields.InputDataFields.
                                    groundtruth_instance_masks])
    image, masks = self.execute_cpu(graph_fn, [])
    self.assertAllEqual(image.shape, [20, 20, 3])
    self.assertAllEqual(masks.shape, [2, 20, 20])
725
726
727
728
729
730
731
732
733
734
735
736
737

  def test_include_keypoints_in_data_augmentation(self):
    data_augmentation_options = [
        (preprocessor.resize_image, {
            'new_height': 20,
            'new_width': 20,
            'method': tf.image.ResizeMethod.NEAREST_NEIGHBOR
        }),
        (preprocessor.scale_boxes_to_pixel_coordinates, {}),
    ]
    data_augmentation_fn = functools.partial(
        inputs.augment_input_data,
        data_augmentation_options=data_augmentation_options)
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(10, 10, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1., 1.]], np.float32)),
          fields.InputDataFields.groundtruth_keypoints:
              tf.constant(np.array([[[0.5, 1.0], [0.5, 0.5]]], np.float32))
      }
      augmented_tensor_dict = data_augmentation_fn(tensor_dict=tensor_dict)
      return (augmented_tensor_dict[fields.InputDataFields.image],
              augmented_tensor_dict[fields.InputDataFields.groundtruth_boxes],
              augmented_tensor_dict[fields.InputDataFields.
                                    groundtruth_keypoints])
    image, boxes, keypoints = self.execute_cpu(graph_fn, [])
    self.assertAllEqual(image.shape, [20, 20, 3])
    self.assertAllClose(boxes, [[10, 10, 20, 20]])
    self.assertAllClose(keypoints, [[[10, 20], [10, 10]]])
756
757
758
759
760
761
762
763
764
765


def _fake_model_preprocessor_fn(image):
  return (image, tf.expand_dims(tf.shape(image)[1:], axis=0))


def _fake_image_resizer_fn(image, mask):
  return (image, mask, tf.shape(image))


766
767
768
769
770
771
772
773
def _fake_resize50_preprocess_fn(image):
  image = image[0]
  image, shape = preprocessor.resize_to_range(
      image, min_dimension=50, max_dimension=50, pad_to_max_dimension=True)

  return tf.expand_dims(image, 0), tf.expand_dims(shape, axis=0)


774
class DataTransformationFnTest(test_case.TestCase, parameterized.TestCase):
775

776
777
778
  def test_combine_additional_channels_if_present(self):
    image = np.random.rand(4, 4, 3).astype(np.float32)
    additional_channels = np.random.rand(4, 4, 2).astype(np.float32)
779
780
781
782
783
784
785
    def graph_fn(image, additional_channels):
      tensor_dict = {
          fields.InputDataFields.image: image,
          fields.InputDataFields.image_additional_channels: additional_channels,
          fields.InputDataFields.groundtruth_classes:
              tf.constant([1, 1], tf.int32)
      }
786

787
788
789
790
791
792
793
794
795
796
797
798
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=1)
      out_tensors = input_transformation_fn(tensor_dict=tensor_dict)
      return out_tensors[fields.InputDataFields.image]
    out_image = self.execute_cpu(graph_fn, [image, additional_channels])
    self.assertAllEqual(out_image.dtype, tf.float32)
    self.assertAllEqual(out_image.shape, [4, 4, 5])
    self.assertAllClose(out_image, np.concatenate((image, additional_channels),
                                                  axis=2))
799

pkulzc's avatar
pkulzc committed
800
  def test_use_multiclass_scores_when_present(self):
801
802
803
804
805
806
807
808
809
810
811
812
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image: tf.constant(np.random.rand(4, 4, 3).
                                                    astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1, 1], [.5, .5, 1, 1]],
                                   np.float32)),
          fields.InputDataFields.multiclass_scores:
              tf.constant(np.array([0.2, 0.3, 0.5, 0.1, 0.6, 0.3], np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([1, 2], np.int32))
      }
pkulzc's avatar
pkulzc committed
813

814
815
816
817
818
819
820
821
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=3, use_multiclass_scores=True)
      transformed_inputs = input_transformation_fn(tensor_dict=tensor_dict)
      return transformed_inputs[fields.InputDataFields.groundtruth_classes]
    groundtruth_classes = self.execute_cpu(graph_fn, [])
pkulzc's avatar
pkulzc committed
822
823
    self.assertAllClose(
        np.array([[0.2, 0.3, 0.5], [0.1, 0.6, 0.3]], np.float32),
824
        groundtruth_classes)
pkulzc's avatar
pkulzc committed
825

826
827
  @unittest.skipIf(tf_version.is_tf2(), ('Skipping due to different behaviour '
                                         'in TF 2.X'))
pkulzc's avatar
pkulzc committed
828
  def test_use_multiclass_scores_when_not_present(self):
829
830
831
832
833
834
835
836
837
838
839
840
841
    def graph_fn():
      zero_num_elements = tf.random.uniform([], minval=0, maxval=1,
                                            dtype=tf.int32)
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1, 1], [.5, .5, 1, 1]],
                                   np.float32)),
          fields.InputDataFields.multiclass_scores: tf.zeros(zero_num_elements),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([1, 2], np.int32))
      }
pkulzc's avatar
pkulzc committed
842

843
844
845
846
847
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=3, use_multiclass_scores=True)
pkulzc's avatar
pkulzc committed
848

849
850
851
      transformed_inputs = input_transformation_fn(tensor_dict=tensor_dict)
      return transformed_inputs[fields.InputDataFields.groundtruth_classes]
    groundtruth_classes = self.execute_cpu(graph_fn, [])
pkulzc's avatar
pkulzc committed
852
853
    self.assertAllClose(
        np.array([[0, 1, 0], [0, 0, 1]], np.float32),
854
        groundtruth_classes)
pkulzc's avatar
pkulzc committed
855

856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
  @parameterized.parameters(
      {'labeled_classes': [1, 2]},
      {'labeled_classes': []},
      {'labeled_classes': [1, -1, 2]}  # -1 denotes an unrecognized class
  )
  def test_use_labeled_classes(self, labeled_classes):

    def compute_fn(image, groundtruth_boxes, groundtruth_classes,
                   groundtruth_labeled_classes):
      tensor_dict = {
          fields.InputDataFields.image:
              image,
          fields.InputDataFields.groundtruth_boxes:
              groundtruth_boxes,
          fields.InputDataFields.groundtruth_classes:
              groundtruth_classes,
          fields.InputDataFields.groundtruth_labeled_classes:
              groundtruth_labeled_classes
      }

      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=3)
      return input_transformation_fn(tensor_dict=tensor_dict)

    image = np.random.rand(4, 4, 3).astype(np.float32)
    groundtruth_boxes = np.array([[.5, .5, 1, 1], [.5, .5, 1, 1]], np.float32)
    groundtruth_classes = np.array([1, 2], np.int32)
    groundtruth_labeled_classes = np.array(labeled_classes, np.int32)

    transformed_inputs = self.execute_cpu(compute_fn, [
        image, groundtruth_boxes, groundtruth_classes,
        groundtruth_labeled_classes
    ])

    if labeled_classes == [1, 2] or labeled_classes == [1, -1, 2]:
      transformed_labeled_classes = [1, 1, 0]
    elif not labeled_classes:
      transformed_labeled_classes = [1, 1, 1]
    else:
      logging.exception('Unexpected labeled_classes %r', labeled_classes)

    self.assertAllEqual(
        np.array(transformed_labeled_classes, np.float32),
        transformed_inputs[fields.InputDataFields.groundtruth_labeled_classes])

904
  def test_returns_correct_class_label_encodings(self):
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[0, 0, 1, 1], [.5, .5, 1, 1]], np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, 1], np.int32))
      }
      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes)
      transformed_inputs = input_transformation_fn(tensor_dict=tensor_dict)
      return (transformed_inputs[fields.InputDataFields.groundtruth_classes],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_confidences])
    (groundtruth_classes, groundtruth_confidences) = self.execute_cpu(graph_fn,
                                                                      [])
    self.assertAllClose(groundtruth_classes, [[0, 0, 1], [1, 0, 0]])
    self.assertAllClose(groundtruth_confidences, [[0, 0, 1], [1, 0, 0]])
928

929
  def test_returns_correct_labels_with_unrecognized_class(self):
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(
                  np.array([[0, 0, 1, 1], [.2, .2, 4, 4], [.5, .5, 1, 1]],
                           np.float32)),
          fields.InputDataFields.groundtruth_area:
              tf.constant(np.array([.5, .4, .3])),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, -1, 1], np.int32)),
          fields.InputDataFields.groundtruth_keypoints:
              tf.constant(
                  np.array([[[.1, .1]], [[.2, .2]], [[.5, .5]]],
                           np.float32)),
          fields.InputDataFields.groundtruth_keypoint_visibilities:
              tf.constant([[True, True], [False, False], [True, True]]),
          fields.InputDataFields.groundtruth_instance_masks:
              tf.constant(np.random.rand(3, 4, 4).astype(np.float32)),
          fields.InputDataFields.groundtruth_is_crowd:
              tf.constant([False, True, False]),
          fields.InputDataFields.groundtruth_difficult:
              tf.constant(np.array([0, 0, 1], np.int32))
      }
955

956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes)
      transformed_inputs = input_transformation_fn(tensor_dict)
      return (transformed_inputs[fields.InputDataFields.groundtruth_classes],
              transformed_inputs[fields.InputDataFields.num_groundtruth_boxes],
              transformed_inputs[fields.InputDataFields.groundtruth_area],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_confidences],
              transformed_inputs[fields.InputDataFields.groundtruth_boxes],
              transformed_inputs[fields.InputDataFields.groundtruth_keypoints],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_keypoint_visibilities],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_instance_masks],
              transformed_inputs[fields.InputDataFields.groundtruth_is_crowd],
              transformed_inputs[fields.InputDataFields.groundtruth_difficult])
    (groundtruth_classes, num_groundtruth_boxes, groundtruth_area,
     groundtruth_confidences, groundtruth_boxes, groundtruth_keypoints,
     groundtruth_keypoint_visibilities, groundtruth_instance_masks,
     groundtruth_is_crowd, groundtruth_difficult) = self.execute_cpu(graph_fn,
                                                                     [])

    self.assertAllClose(groundtruth_classes, [[0, 0, 1], [1, 0, 0]])
    self.assertAllEqual(num_groundtruth_boxes, 2)
    self.assertAllClose(groundtruth_area, [.5, .3])
    self.assertAllEqual(groundtruth_confidences, [[0, 0, 1], [1, 0, 0]])
    self.assertAllClose(groundtruth_boxes, [[0, 0, 1, 1], [.5, .5, 1, 1]])
    self.assertAllClose(groundtruth_keypoints, [[[.1, .1]], [[.5, .5]]])
    self.assertAllEqual(groundtruth_keypoint_visibilities,
                        [[True, True], [True, True]])
    self.assertAllEqual(groundtruth_instance_masks.shape, [2, 4, 4])
    self.assertAllEqual(groundtruth_is_crowd, [False, False])
    self.assertAllEqual(groundtruth_difficult, [0, 1])
993

994
  def test_returns_correct_merged_boxes(self):
995
996
997
998
999
1000
1001
1002
1003
1004
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1, 1], [.5, .5, 1, 1]],
                                   np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, 1], np.int32))
      }
1005

1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes,
          merge_multiple_boxes=True)
      transformed_inputs = input_transformation_fn(tensor_dict)
      return (transformed_inputs[fields.InputDataFields.groundtruth_boxes],
              transformed_inputs[fields.InputDataFields.groundtruth_classes],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_confidences],
              transformed_inputs[fields.InputDataFields.num_groundtruth_boxes])
    (groundtruth_boxes, groundtruth_classes, groundtruth_confidences,
     num_groundtruth_boxes) = self.execute_cpu(graph_fn, [])
1021
    self.assertAllClose(
1022
        groundtruth_boxes,
1023
1024
        [[.5, .5, 1., 1.]])
    self.assertAllClose(
1025
        groundtruth_classes,
1026
        [[1, 0, 1]])
1027
    self.assertAllClose(
1028
        groundtruth_confidences,
1029
        [[1, 0, 1]])
1030
    self.assertAllClose(
1031
        num_groundtruth_boxes,
1032
        1)
1033

1034
  def test_returns_correct_groundtruth_confidences_when_input_present(self):
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[0, 0, 1, 1], [.5, .5, 1, 1]], np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, 1], np.int32)),
          fields.InputDataFields.groundtruth_confidences:
              tf.constant(np.array([1.0, -1.0], np.float32))
      }
      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes)
      transformed_inputs = input_transformation_fn(tensor_dict)
      return (transformed_inputs[fields.InputDataFields.groundtruth_classes],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_confidences])
    groundtruth_classes, groundtruth_confidences = self.execute_cpu(graph_fn,
                                                                    [])
1058
    self.assertAllClose(
1059
        groundtruth_classes,
1060
1061
        [[0, 0, 1], [1, 0, 0]])
    self.assertAllClose(
1062
        groundtruth_confidences,
1063
1064
        [[0, 0, 1], [-1, 0, 0]])

1065
  def test_returns_resized_masks(self):
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(4, 4, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_instance_masks:
              tf.constant(np.random.rand(2, 4, 4).astype(np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, 1], np.int32)),
          fields.InputDataFields.original_image_spatial_shape:
              tf.constant(np.array([4, 4], np.int32))
      }
1077

1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
      def fake_image_resizer_fn(image, masks=None):
        resized_image = tf.image.resize_images(image, [8, 8])
        results = [resized_image]
        if masks is not None:
          resized_masks = tf.transpose(
              tf.image.resize_images(tf.transpose(masks, [1, 2, 0]), [8, 8]),
              [2, 0, 1])
          results.append(resized_masks)
        results.append(tf.shape(resized_image))
        return results

      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=fake_image_resizer_fn,
          num_classes=num_classes,
          retain_original_image=True)
      transformed_inputs = input_transformation_fn(tensor_dict)
      return (transformed_inputs[fields.InputDataFields.original_image],
              transformed_inputs[fields.InputDataFields.
                                 original_image_spatial_shape],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_instance_masks])
    (original_image, original_image_shape,
     groundtruth_instance_masks) = self.execute_cpu(graph_fn, [])
    self.assertEqual(original_image.dtype, np.uint8)
    self.assertAllEqual(original_image_shape, [4, 4])
    self.assertAllEqual(original_image.shape, [8, 8, 3])
    self.assertAllEqual(groundtruth_instance_masks.shape, [2, 8, 8])
1108
1109
1110

  def test_applies_model_preprocess_fn_to_image_tensor(self):
    np_image = np.random.randint(256, size=(4, 4, 3))
1111
1112
1113
1114
1115
1116
    def graph_fn(image):
      tensor_dict = {
          fields.InputDataFields.image: image,
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, 1], np.int32))
      }
1117

1118
1119
      def fake_model_preprocessor_fn(image):
        return (image / 255., tf.expand_dims(tf.shape(image)[1:], axis=0))
1120

1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes)
      transformed_inputs = input_transformation_fn(tensor_dict)
      return (transformed_inputs[fields.InputDataFields.image],
              transformed_inputs[fields.InputDataFields.true_image_shape])
    image, true_image_shape = self.execute_cpu(graph_fn, [np_image])
    self.assertAllClose(image, np_image / 255.)
    self.assertAllClose(true_image_shape, [4, 4, 3])
1133
1134
1135

  def test_applies_data_augmentation_fn_to_tensor_dict(self):
    np_image = np.random.randint(256, size=(4, 4, 3))
1136
1137
1138
1139
1140
1141
    def graph_fn(image):
      tensor_dict = {
          fields.InputDataFields.image: image,
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, 1], np.int32))
      }
1142

1143
1144
      def add_one_data_augmentation_fn(tensor_dict):
        return {key: value + 1 for key, value in tensor_dict.items()}
1145

1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
      num_classes = 4
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes,
          data_augmentation_fn=add_one_data_augmentation_fn)
      transformed_inputs = input_transformation_fn(tensor_dict)
      return (transformed_inputs[fields.InputDataFields.image],
              transformed_inputs[fields.InputDataFields.groundtruth_classes])
    image, groundtruth_classes = self.execute_cpu(graph_fn, [np_image])
    self.assertAllEqual(image, np_image + 1)
    self.assertAllEqual(
        groundtruth_classes,
1160
1161
1162
1163
        [[0, 0, 0, 1], [0, 1, 0, 0]])

  def test_applies_data_augmentation_fn_before_model_preprocess_fn(self):
    np_image = np.random.randint(256, size=(4, 4, 3))
1164
1165
1166
1167
1168
1169
    def graph_fn(image):
      tensor_dict = {
          fields.InputDataFields.image: image,
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([3, 1], np.int32))
      }
1170

1171
1172
      def mul_two_model_preprocessor_fn(image):
        return (image * 2, tf.expand_dims(tf.shape(image)[1:], axis=0))
1173

1174
1175
1176
      def add_five_to_image_data_augmentation_fn(tensor_dict):
        tensor_dict[fields.InputDataFields.image] += 5
        return tensor_dict
1177

1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
      num_classes = 4
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=mul_two_model_preprocessor_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes,
          data_augmentation_fn=add_five_to_image_data_augmentation_fn)
      transformed_inputs = input_transformation_fn(tensor_dict)
      return transformed_inputs[fields.InputDataFields.image]
    image = self.execute_cpu(graph_fn, [np_image])
    self.assertAllEqual(image, (np_image + 5) * 2)
1189

1190
  def test_resize_with_padding(self):
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(100, 50, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1, 1], [.0, .0, .5, .5]],
                                   np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([1, 2], np.int32)),
          fields.InputDataFields.groundtruth_keypoints:
              tf.constant([[[0.1, 0.2]], [[0.3, 0.4]]]),
      }
1203

1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_resize50_preprocess_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes,)
      transformed_inputs = input_transformation_fn(tensor_dict)
      return (transformed_inputs[fields.InputDataFields.groundtruth_boxes],
              transformed_inputs[fields.InputDataFields.groundtruth_keypoints])
    groundtruth_boxes, groundtruth_keypoints = self.execute_cpu(graph_fn, [])
1214
    self.assertAllClose(
1215
        groundtruth_boxes,
1216
1217
        [[.5, .25, 1., .5], [.0, .0, .5, .25]])
    self.assertAllClose(
1218
        groundtruth_keypoints,
1219
1220
1221
        [[[.1, .1]], [[.3, .2]]])

  def test_groundtruth_keypoint_weights(self):
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(100, 50, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1, 1], [.0, .0, .5, .5]],
                                   np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([1, 2], np.int32)),
          fields.InputDataFields.groundtruth_keypoints:
              tf.constant([[[0.1, 0.2], [0.3, 0.4]],
                           [[0.5, 0.6], [0.7, 0.8]]]),
          fields.InputDataFields.groundtruth_keypoint_visibilities:
              tf.constant([[True, False], [True, True]]),
      }
1237

1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
      num_classes = 3
      keypoint_type_weight = [1.0, 2.0]
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_resize50_preprocess_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes,
          keypoint_type_weight=keypoint_type_weight)
      transformed_inputs = input_transformation_fn(tensor_dict=tensor_dict)
      return (transformed_inputs[fields.InputDataFields.groundtruth_keypoints],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_keypoint_weights])

    groundtruth_keypoints, groundtruth_keypoint_weights = self.execute_cpu(
        graph_fn, [])
1253
    self.assertAllClose(
1254
        groundtruth_keypoints,
1255
1256
1257
        [[[0.1, 0.1], [0.3, 0.2]],
         [[0.5, 0.3], [0.7, 0.4]]])
    self.assertAllClose(
1258
        groundtruth_keypoint_weights,
1259
1260
1261
        [[1.0, 0.0], [1.0, 2.0]])

  def test_groundtruth_keypoint_weights_default(self):
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(100, 50, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1, 1], [.0, .0, .5, .5]],
                                   np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([1, 2], np.int32)),
          fields.InputDataFields.groundtruth_keypoints:
              tf.constant([[[0.1, 0.2], [0.3, 0.4]],
                           [[0.5, 0.6], [0.7, 0.8]]]),
      }
1275

1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
      num_classes = 3
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_resize50_preprocess_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes)
      transformed_inputs = input_transformation_fn(tensor_dict=tensor_dict)
      return (transformed_inputs[fields.InputDataFields.groundtruth_keypoints],
              transformed_inputs[fields.InputDataFields.
                                 groundtruth_keypoint_weights])
    groundtruth_keypoints, groundtruth_keypoint_weights = self.execute_cpu(
        graph_fn, [])
1288
    self.assertAllClose(
1289
        groundtruth_keypoints,
1290
1291
1292
        [[[0.1, 0.1], [0.3, 0.2]],
         [[0.5, 0.3], [0.7, 0.4]]])
    self.assertAllClose(
1293
        groundtruth_keypoint_weights,
1294
        [[1.0, 1.0], [1.0, 1.0]])
1295

1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
  def test_groundtruth_dense_pose(self):
    def graph_fn():
      tensor_dict = {
          fields.InputDataFields.image:
              tf.constant(np.random.rand(100, 50, 3).astype(np.float32)),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant(np.array([[.5, .5, 1, 1], [.0, .0, .5, .5]],
                                   np.float32)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant(np.array([1, 2], np.int32)),
          fields.InputDataFields.groundtruth_dp_num_points:
              tf.constant([0, 2], dtype=tf.int32),
          fields.InputDataFields.groundtruth_dp_part_ids:
              tf.constant([[0, 0], [4, 23]], dtype=tf.int32),
          fields.InputDataFields.groundtruth_dp_surface_coords:
              tf.constant([[[0., 0., 0., 0.,], [0., 0., 0., 0.,]],
                           [[0.1, 0.2, 0.3, 0.4,], [0.6, 0.8, 0.6, 0.7,]]],
                          dtype=tf.float32),
      }

      num_classes = 1
      input_transformation_fn = functools.partial(
          inputs.transform_input_data,
          model_preprocess_fn=_fake_resize50_preprocess_fn,
          image_resizer_fn=_fake_image_resizer_fn,
          num_classes=num_classes)
      transformed_inputs = input_transformation_fn(tensor_dict=tensor_dict)
      transformed_dp_num_points = transformed_inputs[
          fields.InputDataFields.groundtruth_dp_num_points]
      transformed_dp_part_ids = transformed_inputs[
          fields.InputDataFields.groundtruth_dp_part_ids]
      transformed_dp_surface_coords = transformed_inputs[
          fields.InputDataFields.groundtruth_dp_surface_coords]
      return (transformed_dp_num_points, transformed_dp_part_ids,
              transformed_dp_surface_coords)

    dp_num_points, dp_part_ids, dp_surface_coords = self.execute_cpu(
        graph_fn, [])
    self.assertAllEqual(dp_num_points, [0, 2])
    self.assertAllEqual(dp_part_ids, [[0, 0], [4, 23]])
    self.assertAllClose(
        dp_surface_coords,
        [[[0., 0., 0., 0.,], [0., 0., 0., 0.,]],
         [[0.1, 0.1, 0.3, 0.4,], [0.6, 0.4, 0.6, 0.7,]]])

1341

pkulzc's avatar
pkulzc committed
1342
class PadInputDataToStaticShapesFnTest(test_case.TestCase):
1343
1344
1345
1346

  def test_pad_images_boxes_and_classes(self):
    input_tensor_dict = {
        fields.InputDataFields.image:
1347
            tf.random.uniform([3, 3, 3]),
1348
        fields.InputDataFields.groundtruth_boxes:
1349
            tf.random.uniform([2, 4]),
1350
        fields.InputDataFields.groundtruth_classes:
1351
            tf.random.uniform([2, 3], minval=0, maxval=2, dtype=tf.int32),
pkulzc's avatar
pkulzc committed
1352
        fields.InputDataFields.true_image_shape:
1353
            tf.constant([3, 3, 3]),
pkulzc's avatar
pkulzc committed
1354
        fields.InputDataFields.original_image_spatial_shape:
1355
            tf.constant([3, 3])
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
    }
    padded_tensor_dict = inputs.pad_input_data_to_static_shapes(
        tensor_dict=input_tensor_dict,
        max_num_boxes=3,
        num_classes=3,
        spatial_image_shape=[5, 6])

    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.image].shape.as_list(),
        [5, 6, 3])
    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.true_image_shape]
        .shape.as_list(), [3])
pkulzc's avatar
pkulzc committed
1369
1370
1371
    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.original_image_spatial_shape]
        .shape.as_list(), [2])
1372
1373
1374
1375
1376
1377
    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.groundtruth_boxes]
        .shape.as_list(), [3, 4])
    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.groundtruth_classes]
        .shape.as_list(), [3, 3])
1378
1379

  def test_clip_boxes_and_classes(self):
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
    def graph_fn():
      input_tensor_dict = {
          fields.InputDataFields.groundtruth_boxes:
              tf.random.uniform([5, 4]),
          fields.InputDataFields.groundtruth_classes:
              tf.random.uniform([2, 3], maxval=10, dtype=tf.int32),
          fields.InputDataFields.num_groundtruth_boxes:
              tf.constant(5)
      }
      padded_tensor_dict = inputs.pad_input_data_to_static_shapes(
          tensor_dict=input_tensor_dict,
          max_num_boxes=3,
          num_classes=3,
          spatial_image_shape=[5, 6])
      return (padded_tensor_dict[fields.InputDataFields.groundtruth_boxes],
              padded_tensor_dict[fields.InputDataFields.groundtruth_classes],
              padded_tensor_dict[fields.InputDataFields.num_groundtruth_boxes])
    (groundtruth_boxes, groundtruth_classes,
     num_groundtruth_boxes) = self.execute_cpu(graph_fn, [])
    self.assertAllEqual(groundtruth_boxes.shape, [3, 4])
    self.assertAllEqual(groundtruth_classes.shape, [3, 3])
    self.assertEqual(num_groundtruth_boxes, 3)
1402
1403
1404
1405

  def test_images_and_additional_channels(self):
    input_tensor_dict = {
        fields.InputDataFields.image:
1406
            test_utils.image_with_dynamic_shape(4, 3, 5),
1407
        fields.InputDataFields.image_additional_channels:
1408
            test_utils.image_with_dynamic_shape(4, 3, 2),
1409
1410
1411
1412
1413
1414
1415
    }
    padded_tensor_dict = inputs.pad_input_data_to_static_shapes(
        tensor_dict=input_tensor_dict,
        max_num_boxes=3,
        num_classes=3,
        spatial_image_shape=[5, 6])

1416
1417
    # pad_input_data_to_static_shape assumes that image is already concatenated
    # with additional channels.
1418
1419
1420
1421
1422
1423
1424
    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.image].shape.as_list(),
        [5, 6, 5])
    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.image_additional_channels]
        .shape.as_list(), [5, 6, 2])

1425
1426
1427
  def test_images_and_additional_channels_errors(self):
    input_tensor_dict = {
        fields.InputDataFields.image:
1428
            test_utils.image_with_dynamic_shape(10, 10, 3),
1429
        fields.InputDataFields.image_additional_channels:
1430
            test_utils.image_with_dynamic_shape(10, 10, 2),
1431
        fields.InputDataFields.original_image:
1432
            test_utils.image_with_dynamic_shape(10, 10, 3),
1433
1434
1435
1436
1437
1438
1439
1440
    }
    with self.assertRaises(ValueError):
      _ = inputs.pad_input_data_to_static_shapes(
          tensor_dict=input_tensor_dict,
          max_num_boxes=3,
          num_classes=3,
          spatial_image_shape=[5, 6])

1441
1442
1443
  def test_gray_images(self):
    input_tensor_dict = {
        fields.InputDataFields.image:
1444
            test_utils.image_with_dynamic_shape(4, 4, 1),
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
    }
    padded_tensor_dict = inputs.pad_input_data_to_static_shapes(
        tensor_dict=input_tensor_dict,
        max_num_boxes=3,
        num_classes=3,
        spatial_image_shape=[5, 6])

    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.image].shape.as_list(),
        [5, 6, 1])

  def test_gray_images_and_additional_channels(self):
    input_tensor_dict = {
        fields.InputDataFields.image:
1459
            test_utils.image_with_dynamic_shape(4, 4, 3),
1460
        fields.InputDataFields.image_additional_channels:
1461
            test_utils.image_with_dynamic_shape(4, 4, 2),
1462
    }
1463
1464
    # pad_input_data_to_static_shape assumes that image is already concatenated
    # with additional channels.
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
    padded_tensor_dict = inputs.pad_input_data_to_static_shapes(
        tensor_dict=input_tensor_dict,
        max_num_boxes=3,
        num_classes=3,
        spatial_image_shape=[5, 6])

    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.image].shape.as_list(),
        [5, 6, 3])
    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.image_additional_channels]
        .shape.as_list(), [5, 6, 2])

1478
  def test_keypoints(self):
1479
1480
1481
    keypoints = test_utils.keypoints_with_dynamic_shape(10, 16, 4)
    visibilities = tf.cast(tf.random.uniform(tf.shape(keypoints)[:-1], minval=0,
                                             maxval=2, dtype=tf.int32), tf.bool)
1482
1483
    input_tensor_dict = {
        fields.InputDataFields.groundtruth_keypoints:
1484
            test_utils.keypoints_with_dynamic_shape(10, 16, 4),
1485
        fields.InputDataFields.groundtruth_keypoint_visibilities:
1486
            visibilities
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
    }
    padded_tensor_dict = inputs.pad_input_data_to_static_shapes(
        tensor_dict=input_tensor_dict,
        max_num_boxes=3,
        num_classes=3,
        spatial_image_shape=[5, 6])

    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.groundtruth_keypoints]
        .shape.as_list(), [3, 16, 4])
    self.assertAllEqual(
        padded_tensor_dict[
            fields.InputDataFields.groundtruth_keypoint_visibilities]
        .shape.as_list(), [3, 16])

1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
  def test_dense_pose(self):
    input_tensor_dict = {
        fields.InputDataFields.groundtruth_dp_num_points:
            tf.constant([0, 2], dtype=tf.int32),
        fields.InputDataFields.groundtruth_dp_part_ids:
            tf.constant([[0, 0], [4, 23]], dtype=tf.int32),
        fields.InputDataFields.groundtruth_dp_surface_coords:
            tf.constant([[[0., 0., 0., 0.,], [0., 0., 0., 0.,]],
                         [[0.1, 0.2, 0.3, 0.4,], [0.6, 0.8, 0.6, 0.7,]]],
                        dtype=tf.float32),
    }

    padded_tensor_dict = inputs.pad_input_data_to_static_shapes(
        tensor_dict=input_tensor_dict,
        max_num_boxes=3,
        num_classes=1,
        spatial_image_shape=[128, 128],
        max_dp_points=200)

    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.groundtruth_dp_num_points]
        .shape.as_list(), [3])
    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.groundtruth_dp_part_ids]
        .shape.as_list(), [3, 200])
    self.assertAllEqual(
        padded_tensor_dict[fields.InputDataFields.groundtruth_dp_surface_coords]
        .shape.as_list(), [3, 200, 4])

1531
1532
1533
1534
  def test_context_features(self):
    context_memory_size = 8
    context_feature_length = 10
    max_num_context_features = 20
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
    def graph_fn():
      input_tensor_dict = {
          fields.InputDataFields.context_features:
              tf.ones([context_memory_size, context_feature_length]),
          fields.InputDataFields.context_feature_length:
              tf.constant(context_feature_length)
      }
      padded_tensor_dict = inputs.pad_input_data_to_static_shapes(
          tensor_dict=input_tensor_dict,
          max_num_boxes=3,
          num_classes=3,
          spatial_image_shape=[5, 6],
          max_num_context_features=max_num_context_features,
          context_feature_length=context_feature_length)
1549

1550
1551
1552
1553
1554
      self.assertAllEqual(
          padded_tensor_dict[
              fields.InputDataFields.context_features].shape.as_list(),
          [max_num_context_features, context_feature_length])
      return padded_tensor_dict[fields.InputDataFields.valid_context_size]
1555

1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
    valid_context_size = self.execute_cpu(graph_fn, [])
    self.assertEqual(valid_context_size, context_memory_size)


class NegativeSizeTest(test_case.TestCase):
  """Test for inputs and related funcitons."""

  def test_negative_size_error(self):
    """Test that error is raised for negative size boxes."""

    def graph_fn():
      tensors = {
          fields.InputDataFields.image: tf.zeros((128, 128, 3)),
          fields.InputDataFields.groundtruth_classes:
              tf.constant([1, 1], tf.int32),
          fields.InputDataFields.groundtruth_boxes:
              tf.constant([[0.5, 0.5, 0.4, 0.5]], tf.float32)
1573
      }
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
      tensors = inputs.transform_input_data(
          tensors, _fake_model_preprocessor_fn, _fake_image_resizer_fn,
          num_classes=10)
      return tensors[fields.InputDataFields.groundtruth_boxes]
    with self.assertRaises(tf.errors.InvalidArgumentError):
      self.execute_cpu(graph_fn, [])

  def test_negative_size_no_assert(self):
    """Test that negative size boxes are filtered out without assert.

    This test simulates the behaviour when we run on TPU and Assert ops are
    not supported.
    """
1587

1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
    tensors = {
        fields.InputDataFields.image: tf.zeros((128, 128, 3)),
        fields.InputDataFields.groundtruth_classes:
            tf.constant([1, 1], tf.int32),
        fields.InputDataFields.groundtruth_boxes:
            tf.constant([[0.5, 0.5, 0.4, 0.5], [0.5, 0.5, 0.6, 0.6]],
                        tf.float32)
    }

    with mock.patch.object(tf, 'Assert') as tf_assert:
      tf_assert.return_value = tf.no_op()
      tensors = inputs.transform_input_data(
          tensors, _fake_model_preprocessor_fn, _fake_image_resizer_fn,
          num_classes=10)

      self.assertAllClose(tensors[fields.InputDataFields.groundtruth_boxes],
                          [[0.5, 0.5, 0.6, 0.6]])
1605

1606

1607
1608
if __name__ == '__main__':
  tf.test.main()