def default_classification_model(
num_classes,
num_anchors,
pyramid_feature_size=256,
prior_probability=0.01,
classification_feature_size=256,
name='classification_submodel'
):
""" Creates the default classification submodel.
Args
num_classes : Number of classes to predict a score for at each feature level.
num_anchors : Number of anchors to predict classification scores for at each feature level.
pyramid_feature_size : The number of filters to expect from the feature pyramid levels.
classification_feature_size : The number of filters to use in the layers in the classification submodel.
name : The name of the submodel.
Returns
A keras.models.Model that predicts classes for each anchor.
"""
options = {
'kernel_size' : 3,
'strides' : 1,
'padding' : 'same',
}
if keras.backend.image_data_format() == 'channels_first':
inputs = keras.layers.Input(shape=(pyramid_feature_size, None, None))
else:
inputs = keras.layers.Input(shape=(None, None, pyramid_feature_size))
outputs = inputs
for i in range(4):
outputs = keras.layers.Conv2D(
filters=classification_feature_size,
activation='relu',
name='pyramid_classification_{}'.format(i),
kernel_initializer=keras.initializers.normal(mean=0.0, stddev=0.01, seed=None),
bias_initializer='zeros',
**options
)(outputs)
outputs = keras.layers.Conv2D(
filters=num_classes * num_anchors,
kernel_initializer=keras.initializers.normal(mean=0.0, stddev=0.01, seed=None),
bias_initializer=initializers.PriorProbability(probability=prior_probability),
name='pyramid_classification',
**options
)(outputs)
# reshape output and apply sigmoid
if keras.backend.image_data_format() == 'channels_first':
outputs = keras.layers.Permute((2, 3, 1), name='pyramid_classification_permute')(outputs)
outputs = keras.layers.Reshape((-1, num_classes), name='pyramid_classification_reshape')(outputs)
outputs = keras.layers.Activation('sigmoid', name='pyramid_classification_sigmoid')(outputs)
return keras.models.Model(inputs=inputs, outputs=outputs, name=name)
def default_classification_model(num_classes,
shared_model,
pyramid_feature_size=256,
prior_probability=0.01,
name='classification_submodel'):
"""
Creates the default classification submodel.
Args
num_classes: Number of classes to predict a score for at each feature level.
shared_model:
pyramid_feature_size: The number of filters to expect from the feature pyramid levels.
name: The name of the submodel.
Returns
A keras.models.Model that predicts classes for each anchor.
"""
options = {
'kernel_size': 3,
'strides': 1,
'padding': 'same',
}
inputs = keras.layers.Input(shape=(None, None, pyramid_feature_size))
outputs = shared_model(inputs)
outputs = keras.layers.Conv2D(
filters=num_classes,
kernel_initializer=keras.initializers.normal(mean=0.0,
stddev=0.01,
seed=None),
bias_initializer=initializers.PriorProbability(
probability=prior_probability),
name='pyramid_classification',
**options)(outputs)
# reshape output and apply sigmoid
outputs = keras.layers.Reshape(
(-1, num_classes), name='pyramid_classification_reshape')(outputs)
outputs = keras.layers.Activation(
'sigmoid', name='pyramid_classification_sigmoid')(outputs)
return keras.models.Model(inputs=inputs, outputs=outputs, name=name)
def default_classification_model(num_classes,
pyramid_feature_size=256,
prior_probability=0.01,
classification_feature_size=256,
name='classification_submodel'):
options = {
'kernel_size': 3,
'strides': 1,
'padding': 'same',
}
inputs = keras.layers.Input(shape=(None, None, pyramid_feature_size))
outputs = inputs
for i in range(4):
outputs = keras.layers.Conv2D(
filters=classification_feature_size,
activation='relu',
name='pyramid_classification_{}'.format(i),
kernel_initializer=keras.initializers.normal(mean=0.0,
stddev=0.01,
seed=None),
bias_initializer='zeros',
**options)(outputs)
outputs = keras.layers.Conv2D(
filters=num_classes,
kernel_initializer=keras.initializers.zeros(),
bias_initializer=initializers.PriorProbability(
probability=prior_probability),
name='pyramid_classification',
**options)(outputs)
# reshape output and apply sigmoid
outputs = keras.layers.Reshape(
(-1, num_classes), name='pyramid_classification_reshape')(outputs)
outputs = keras.layers.Activation(
'sigmoid', name='pyramid_classification_sigmoid')(outputs)
return keras.models.Model(inputs=inputs, outputs=outputs, name=name)
def default_classification_model(num_classes,
num_anchors,
pyramid_feature_size=256,
name='classification_submodel',
width=256,
depth=1,
separable_conv=False,
kernel_size=3):
options = {
'kernel_size': kernel_size,
'strides': 1,
'padding': 'same',
}
if keras.backend.image_data_format() == 'channels_first':
inputs = keras.layers.Input(shape=(pyramid_feature_size, None, None))
else:
inputs = keras.layers.Input(shape=(None, None, pyramid_feature_size))
if separable_conv:
kernel_initializer = {
'depthwise_initializer': keras.initializers.VarianceScaling(),
'pointwise_initializer': keras.initializers.VarianceScaling(),
}
options.update(kernel_initializer)
convs = [
keras.layers.SeparableConv2D(filters=width,
bias_initializer='zeros',
name=f'{name}/class-{i}',
**options) for i in range(depth)
]
head = keras.layers.SeparableConv2D(
filters=num_classes * num_anchors,
bias_initializer=initializers.PriorProbability(probability=0.01),
name=f'{name}/class-predict',
**options)
else:
kernel_initializer = {
'kernel_initializer':
keras.initializers.RandomNormal(mean=0.0, stddev=0.01, seed=None)
}
options.update(kernel_initializer)
convs = [
keras.layers.Conv2D(filters=width,
bias_initializer='zeros',
name=f'{name}/class-{i}',
**options) for i in range(depth)
]
head = keras.layers.Conv2D(
filters=num_classes * num_anchors,
bias_initializer=initializers.PriorProbability(probability=0.01),
name='class-predict',
**options)
bns = [[
keras.layers.BatchNormalization(momentum=MOMENTUM,
epsilon=EPSILON,
name=f'{name}/class-{i}-bn-{j}')
for j in range(3, 8)
] for i in range(depth)]
relu = keras.layers.Activation('sigmoid',
name='pyramid_classification_sigmoid')
reshape = keras.layers.Reshape((-1, num_classes))
activation = keras.layers.Activation('sigmoid')
level = 0
outputs = inputs
for i in range(depth):
outputs = convs[i](outputs)
outputs = bns[i][level](outputs)
outputs = relu(outputs)
level += 1
outputs = head(outputs)
outputs = reshape(outputs)
outputs = activation(outputs)
return keras.models.Model(inputs=inputs, outputs=outputs, name=name)