Skip to content

mobilenet

MobileNet Models API

This module provides utility functions to generate MobileNet models.

Classes:

Functions:

Classes

MobileNetV1Params

MobileNetV1 parameters

Attributes:

  • input_filters (int) –

    Input filters

  • input_strides (int | tuple[int, int]) –

    Input stride

  • include_top (bool) –

    Include top

  • output_activation (str | None) –

    Output activation

  • name (str) –

    Model name

MobileNetV1Model

Helper class to generate model from parameters

Functions

layer_from_params staticmethod 
layer_from_params(inputs: keras.Input, params: MobileNetV1Params | dict, num_classes: int | None = None)

Create layer from parameters

Source code in neuralspot_edge/models/mobilenet.py 
296
297
298
299
300
301
@staticmethod
def layer_from_params(inputs: keras.Input, params: MobileNetV1Params | dict, num_classes: int | None = None):
    """Create layer from parameters"""
    if isinstance(params, dict):
        params = MobileNetV1Params(**params)
    return mobilenetv1_layer(x=inputs, params=params, num_classes=num_classes)
model_from_params staticmethod 
model_from_params(inputs: keras.Input, params: MobileNetV1Params | dict, num_classes: int | None = None)

Create model from parameters

Source code in neuralspot_edge/models/mobilenet.py 
303
304
305
306
307
@staticmethod
def model_from_params(inputs: keras.Input, params: MobileNetV1Params | dict, num_classes: int | None = None):
    """Create model from parameters"""
    outputs = MobileNetV1Model.layer_from_params(inputs=inputs, params=params, num_classes=num_classes)
    return keras.Model(inputs=inputs, outputs=outputs)

Functions

mobilenetv1_layer 

mobilenetv1_layer(x: keras.KerasTensor, num_classes: int, params: MobileNetV1Params) -> keras.KerasTensor

Modified MobileNetV1 MLPerf Tiny model for VWW: https://github.com/SiliconLabs/platform_ml_models/blob/master/eembc/Person_detection/mobilenet_v1_eembc.py

Parameters:

  • x (KerasTensor) –

    Model input

  • num_classes (int) –

    Number of classes.

  • params (MobileNetV1Params) –

    Model parameters

Returns:

  • KerasTensor –

    keras.KerasTensor: Model output

Source code in neuralspot_edge/models/mobilenet.py 
 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
227
228
229
230
231
232
233
234
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
def mobilenetv1_layer(
    x: keras.KerasTensor,
    num_classes: int,
    params: MobileNetV1Params,
) -> keras.KerasTensor:
    """Modified MobileNetV1
    MLPerf Tiny model for VWW:
    https://github.com/SiliconLabs/platform_ml_models/blob/master/eembc/Person_detection/mobilenet_v1_eembc.py

    Args:
        x (keras.KerasTensor): Model input
        num_classes (int): Number of classes.
        params (MobileNetV1Params): Model parameters

    Returns:
        keras.KerasTensor: Model output
    """
    num_filters = params.input_filters

    # 1st layer, pure conv
    # Keras 2.2 model has padding='valid' and disables bias
    y = keras.layers.Conv2D(
        num_filters,
        kernel_size=3,
        strides=params.input_strides,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(x)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)  # Keras uses ReLU6 instead of pure ReLU

    # 2nd layer, depthwise separable conv
    # Filter size is always doubled before the pointwise conv
    # Keras uses ZeroPadding2D() and padding='valid'
    y = keras.layers.DepthwiseConv2D(
        kernel_size=3,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    num_filters = 2 * num_filters
    y = keras.layers.Conv2D(
        num_filters,
        kernel_size=1,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    # 3rd layer, depthwise separable conv
    y = keras.layers.DepthwiseConv2D(
        kernel_size=3,
        strides=2,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    num_filters = 2 * num_filters
    y = keras.layers.Conv2D(
        num_filters,
        kernel_size=1,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    # 4th layer, depthwise separable conv
    y = keras.layers.DepthwiseConv2D(
        kernel_size=3,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    y = keras.layers.Conv2D(
        num_filters,
        kernel_size=1,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    # 5th layer, depthwise separable conv
    y = keras.layers.DepthwiseConv2D(
        kernel_size=3,
        strides=2,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    num_filters = 2 * num_filters
    y = keras.layers.Conv2D(
        num_filters,
        kernel_size=1,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    # 6th layer, depthwise separable conv
    y = keras.layers.DepthwiseConv2D(
        kernel_size=3,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    y = keras.layers.Conv2D(
        num_filters,
        kernel_size=1,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    # 7th layer, depthwise separable conv
    y = keras.layers.DepthwiseConv2D(
        kernel_size=3,
        strides=2,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    num_filters = 2 * num_filters
    y = keras.layers.Conv2D(
        num_filters,
        kernel_size=1,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    # 8th-12th layers, identical depthwise separable convs
    for _ in range(8, 13):
        y = keras.layers.DepthwiseConv2D(
            kernel_size=3,
            strides=1,
            padding="same",
            kernel_initializer="he_normal",
            kernel_regularizer=keras.regularizers.l2(1e-4),
        )(y)
        y = keras.layers.BatchNormalization()(y)
        y = keras.layers.Activation("relu")(y)

        y = keras.layers.Conv2D(
            num_filters,
            kernel_size=1,
            strides=1,
            padding="same",
            kernel_initializer="he_normal",
            kernel_regularizer=keras.regularizers.l2(1e-4),
        )(y)
        y = keras.layers.BatchNormalization()(y)
        y = keras.layers.Activation("relu")(y)

    # 13th layer, depthwise separable conv
    y = keras.layers.DepthwiseConv2D(
        kernel_size=3,
        strides=2,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    num_filters = 2 * num_filters
    y = keras.layers.Conv2D(
        num_filters,
        kernel_size=1,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    # 14th layer, depthwise separable conv
    y = keras.layers.DepthwiseConv2D(
        kernel_size=3,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = keras.layers.Activation("relu")(y)

    y = keras.layers.Conv2D(
        num_filters,
        kernel_size=1,
        strides=1,
        padding="same",
        kernel_initializer="he_normal",
        kernel_regularizer=keras.regularizers.l2(1e-4),
    )(y)
    y = keras.layers.BatchNormalization()(y)
    y = cast(keras.KerasTensor, keras.layers.Activation("relu")(y))

    # Average pooling, max polling may be used also
    # Keras employs GlobalAveragePooling2D
    y = keras.layers.AveragePooling2D(pool_size=y.shape[1:3])(y)
    # x = MaxPooling2D(pool_size=x.shape[1:3])(x)

    # Keras inserts Dropout() and a pointwise Conv2D() here
    # We are staying with the paper base structure

    # Flatten, FC layer and classify
    if params.include_top:
        y = keras.layers.Flatten()(y)
        y = keras.layers.Dense(num_classes)(y)

    return y