def build(self, input_shape):
input_shape = tf.TensorShape(input_shape)
self.input_dim = int(input_shape[-1])
self.kernel_shape = [self.input_dim, self.units]
cond_dense_kernel_shape = (self.num_experts, self.input_dim * self.units)
self.cond_dense_kernel = self.add_weight(
name='cond_dense_kernel',
shape=cond_dense_kernel_shape,
initializer=initializers.get_condconv_initializer(
self.kernel_initializer, self.num_experts, self.kernel_shape),
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
trainable=True,
dtype=self.dtype)
if self.use_bias:
self.bias_shape = (self.units,)
cond_dense_bias_shape = (self.num_experts, self.units)
self.cond_dense_bias = self.add_weight(
name='cond_dense_bias',
shape=cond_dense_bias_shape,
initializer=initializers.get_condconv_initializer(
self.bias_initializer, self.num_experts, self.bias_shape),
regularizer=self.bias_regularizer,
constraint=self.bias_constraint,
trainable=True,
dtype=self.dtype)
else:
self.bias = None
self.built = True
def build(self, input_shape):
if len(input_shape) < 4:
raise ValueError(
'Inputs to `DepthwiseCondConv2D` should have rank 4. '
'Received input shape:', str(input_shape))
input_shape = tf.TensorShape(input_shape)
if self.data_format == 'channels_first':
channel_axis = 1
else:
channel_axis = 3
if input_shape.dims[channel_axis].value is None:
raise ValueError('The channel dimension of the inputs to '
'`DepthwiseConv2D` '
'should be defined. Found `None`.')
input_dim = int(input_shape[channel_axis])
self.depthwise_kernel_shape = (self.kernel_size[0], self.kernel_size[1],
input_dim, self.depth_multiplier)
depthwise_kernel_num_params = 1
for dim in self.depthwise_kernel_shape:
depthwise_kernel_num_params *= dim
depthwise_condconv_kernel_shape = (self.num_experts,
depthwise_kernel_num_params)
self.depthwise_condconv_kernel = self.add_weight(
shape=depthwise_condconv_kernel_shape,
initializer=initializers.get_condconv_initializer(
self.depthwise_initializer,
self.num_experts,
self.depthwise_kernel_shape),
name='depthwise_condconv_kernel',
regularizer=self.depthwise_regularizer,
constraint=self.depthwise_constraint,
trainable=True)
if self.use_bias:
bias_dim = input_dim * self.depth_multiplier
self.bias_shape = (bias_dim,)
condconv_bias_shape = (self.num_experts, bias_dim)
self.condconv_bias = self.add_weight(
name='condconv_bias',
shape=condconv_bias_shape,
initializer=initializers.get_condconv_initializer(
self.bias_initializer,
self.num_experts,
self.bias_shape),
regularizer=self.bias_regularizer,
constraint=self.bias_constraint,
trainable=True,
dtype=self.dtype)
else:
self.bias = None
# Set input spec.
self.input_spec = tf.keras.layers.InputSpec(
ndim=4, axes={channel_axis: input_dim})
self.built = True
def build(self, input_shape):
if len(input_shape) != 4:
raise ValueError(
'Inputs to `CondConv2D` should have rank 4. '
'Received input shape:', str(input_shape))
input_shape = tf.TensorShape(input_shape)
channel_axis = self._get_channel_axis()
if input_shape.dims[channel_axis].value is None:
raise ValueError('The channel dimension of the inputs '
'should be defined. Found `None`.')
input_dim = int(input_shape[channel_axis])
self.kernel_shape = self.kernel_size + (input_dim, self.filters)
kernel_num_params = 1
for kernel_dim in self.kernel_shape:
kernel_num_params *= kernel_dim
condconv_kernel_shape = (self.num_experts, kernel_num_params)
self.condconv_kernel = self.add_weight(
name='condconv_kernel',
shape=condconv_kernel_shape,
initializer=initializers.get_condconv_initializer(
self.kernel_initializer,
self.num_experts,
self.kernel_shape),
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
trainable=True,
dtype=self.dtype)
if self.use_bias:
self.bias_shape = (self.filters,)
condconv_bias_shape = (self.num_experts, self.filters)
self.condconv_bias = self.add_weight(
name='condconv_bias',
shape=condconv_bias_shape,
initializer=initializers.get_condconv_initializer(
self.bias_initializer,
self.num_experts,
self.bias_shape),
regularizer=self.bias_regularizer,
constraint=self.bias_constraint,
trainable=True,
dtype=self.dtype)
else:
self.bias = None
self.input_spec = tf.keras.layers.InputSpec(
ndim=self.rank + 2, axes={channel_axis: input_dim})
self.built = True