File size: 7,873 Bytes
5672777
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
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
97
98
99
100
101
102
103
104
105
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
132
133
134
135
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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
# Copyright 2023 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 movinet.py."""

from absl.testing import parameterized
import tensorflow as tf, tf_keras

from official.projects.movinet.modeling import movinet


class MoViNetTest(parameterized.TestCase, tf.test.TestCase):

  def test_network_creation(self):
    """Test creation of MoViNet family models."""
    tf_keras.backend.set_image_data_format('channels_last')

    network = movinet.Movinet(
        model_id='a0',
        causal=True,
    )
    inputs = tf_keras.Input(shape=(8, 128, 128, 3), batch_size=1)
    endpoints, states = network(inputs)

    self.assertAllEqual(endpoints['stem'].shape, [1, 8, 64, 64, 8])
    self.assertAllEqual(endpoints['block0_layer0'].shape, [1, 8, 32, 32, 8])
    self.assertAllEqual(endpoints['block1_layer0'].shape, [1, 8, 16, 16, 32])
    self.assertAllEqual(endpoints['block2_layer0'].shape, [1, 8, 8, 8, 56])
    self.assertAllEqual(endpoints['block3_layer0'].shape, [1, 8, 8, 8, 56])
    self.assertAllEqual(endpoints['block4_layer0'].shape, [1, 8, 4, 4, 104])
    self.assertAllEqual(endpoints['head'].shape, [1, 1, 1, 1, 480])

    self.assertNotEmpty(states)

  def test_network_with_states(self):
    """Test creation of MoViNet family models with states."""
    tf_keras.backend.set_image_data_format('channels_last')

    backbone = movinet.Movinet(
        model_id='a0',
        causal=True,
        use_external_states=True,
    )
    inputs = tf.ones([1, 8, 128, 128, 3])

    init_states = backbone.init_states(tf.shape(inputs))
    endpoints, new_states = backbone({**init_states, 'image': inputs})

    self.assertAllEqual(endpoints['stem'].shape, [1, 8, 64, 64, 8])
    self.assertAllEqual(endpoints['block0_layer0'].shape, [1, 8, 32, 32, 8])
    self.assertAllEqual(endpoints['block1_layer0'].shape, [1, 8, 16, 16, 32])
    self.assertAllEqual(endpoints['block2_layer0'].shape, [1, 8, 8, 8, 56])
    self.assertAllEqual(endpoints['block3_layer0'].shape, [1, 8, 8, 8, 56])
    self.assertAllEqual(endpoints['block4_layer0'].shape, [1, 8, 4, 4, 104])
    self.assertAllEqual(endpoints['head'].shape, [1, 1, 1, 1, 480])

    self.assertNotEmpty(init_states)
    self.assertNotEmpty(new_states)

  def test_movinet_stream(self):
    """Test if the backbone can be run in streaming mode."""
    tf_keras.backend.set_image_data_format('channels_last')

    backbone = movinet.Movinet(
        model_id='a0',
        causal=True,
        use_external_states=True,
    )
    inputs = tf.ones([1, 5, 128, 128, 3])

    init_states = backbone.init_states(tf.shape(inputs))
    expected_endpoints, _ = backbone({**init_states, 'image': inputs})

    frames = tf.split(inputs, inputs.shape[1], axis=1)

    states = init_states
    for frame in frames:
      output, states = backbone({**states, 'image': frame})
    predicted_endpoints = output

    predicted = predicted_endpoints['head']

    # The expected final output is simply the mean across frames
    expected = expected_endpoints['head']
    expected = tf.reduce_mean(expected, 1, keepdims=True)

    self.assertEqual(predicted.shape, expected.shape)
    self.assertAllClose(predicted, expected, 1e-5, 1e-5)

  def test_movinet_stream_nse(self):
    """Test if the backbone can be run in streaming mode w/o SE layer."""
    tf_keras.backend.set_image_data_format('channels_last')

    backbone = movinet.Movinet(
        model_id='a0',
        causal=True,
        use_external_states=True,
        se_type='none',
    )
    inputs = tf.ones([1, 5, 128, 128, 3])

    init_states = backbone.init_states(tf.shape(inputs))
    expected_endpoints, _ = backbone({**init_states, 'image': inputs})

    frames = tf.split(inputs, inputs.shape[1], axis=1)

    states = init_states
    for frame in frames:
      output, states = backbone({**states, 'image': frame})
    predicted_endpoints = output

    predicted = predicted_endpoints['head']

    # The expected final output is simply the mean across frames
    expected = expected_endpoints['head']
    expected = tf.reduce_mean(expected, 1, keepdims=True)

    self.assertEqual(predicted.shape, expected.shape)
    self.assertAllClose(predicted, expected, 1e-5, 1e-5)

    # Check contents in the states dictionary.
    state_keys = list(init_states.keys())
    self.assertIn('state_head_pool_buffer', state_keys)
    self.assertIn('state_head_pool_frame_count', state_keys)
    state_keys.remove('state_head_pool_buffer')
    state_keys.remove('state_head_pool_frame_count')
    # From now on, there are only 'stream_buffer' for the convolutions.
    for state_key in state_keys:
      self.assertIn(
          'stream_buffer', state_key,
          msg=f'Expecting stream_buffer only, found {state_key}')

  def test_movinet_2plus1d_stream(self):
    tf_keras.backend.set_image_data_format('channels_last')

    backbone = movinet.Movinet(
        model_id='a0',
        causal=True,
        conv_type='2plus1d',
        use_external_states=True,
    )
    inputs = tf.ones([1, 5, 128, 128, 3])

    init_states = backbone.init_states(tf.shape(inputs))
    expected_endpoints, _ = backbone({**init_states, 'image': inputs})

    frames = tf.split(inputs, inputs.shape[1], axis=1)

    states = init_states
    for frame in frames:
      output, states = backbone({**states, 'image': frame})
    predicted_endpoints = output

    predicted = predicted_endpoints['head']

    # The expected final output is simply the mean across frames
    expected = expected_endpoints['head']
    expected = tf.reduce_mean(expected, 1, keepdims=True)

    self.assertEqual(predicted.shape, expected.shape)
    self.assertAllClose(predicted, expected, 1e-5, 1e-5)

  def test_movinet_3d_2plus1d_stream(self):
    tf_keras.backend.set_image_data_format('channels_last')

    backbone = movinet.Movinet(
        model_id='a0',
        causal=True,
        conv_type='3d_2plus1d',
        use_external_states=True,
    )
    inputs = tf.ones([1, 5, 128, 128, 3])

    init_states = backbone.init_states(tf.shape(inputs))
    expected_endpoints, _ = backbone({**init_states, 'image': inputs})

    frames = tf.split(inputs, inputs.shape[1], axis=1)

    states = init_states
    for frame in frames:
      output, states = backbone({**states, 'image': frame})
    predicted_endpoints = output

    predicted = predicted_endpoints['head']

    # The expected final output is simply the mean across frames
    expected = expected_endpoints['head']
    expected = tf.reduce_mean(expected, 1, keepdims=True)

    self.assertEqual(predicted.shape, expected.shape)
    self.assertAllClose(predicted, expected, 1e-5, 1e-5)

  def test_serialize_deserialize(self):
    # Create a network object that sets all of its config options.
    kwargs = dict(
        model_id='a0',
        causal=True,
        use_positional_encoding=True,
        use_external_states=True,
    )
    network = movinet.Movinet(**kwargs)

    # Create another network object from the first object's config.
    new_network = movinet.Movinet.from_config(network.get_config())

    # Validate that the config can be forced to JSON.
    _ = new_network.to_json()

    # If the serialization was successful, the new config should match the old.
    self.assertAllEqual(network.get_config(), new_network.get_config())


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