def call(self, inputs, training=None, **kwargs):
layer_util.check_inputs(inputs, 2, "ResidualBlock")
return self._act(
self._norm(inputs=self._conv3d(inputs=inputs[0]), training=training)
+ inputs[1]
)
def build(self, input_shape):
super(LocalNetUpSampleResnetBlock, self).build(input_shape)
layer_util.check_inputs(input_shape, 2, "UpSampleResnetBlock build")
output_shape = input_shape[1][1:4]
self._deconv3d_block = Deconv3dBlock(
filters=self._filters, output_shape=output_shape, strides=2
)
if self._use_additive_upsampling:
self._additive_upsampling = AdditiveUpSampling(output_shape=output_shape)
def call(self, inputs, training=None, **kwargs):
layer_util.check_inputs(inputs, 2, "UpSampleResnetBlock call")
inputs_nonskip, inputs_skip = inputs[0], inputs[1]
h0 = self._deconv3d_block(inputs=inputs_nonskip, training=training)
if self._use_additive_upsampling:
h0 += self._additive_upsampling(inputs=inputs_nonskip)
r1 = h0 + inputs_skip
r2 = self._conv3d_block(inputs=h0, training=training)
h1 = self._residual_block(inputs=[r2, r1], training=training)
return h1
def build(self, input_shape):
"""
:param input_shape: list [nonskip_tensor_shape, skip_tensor_shape]
:return: None
"""
super(LocalNetUpSampleResnetBlock, self).build(input_shape)
layer_util.check_inputs(input_shape, 2, "UpSampleResnetBlock build")
output_shape = input_shape[1][1:4]
self._deconv3d_block = Deconv3dBlock(
filters=self._filters, output_shape=output_shape, strides=2
)
if self._use_additive_upsampling:
self._additive_upsampling = AdditiveUpSampling(output_shape=output_shape)
def call(self, inputs, training=None, **kwargs):
"""
:param inputs: list = [inputs_nonskip, inputs_skip]
:param training (bool): training flag for normalization layers (default: None)
:param kwargs:
:return:
"""
layer_util.check_inputs(inputs, 2, "UpSampleResnetBlock call")
inputs_nonskip, inputs_skip = inputs[0], inputs[1]
h0 = self._deconv3d_block(inputs=inputs_nonskip, training=training)
if self._use_additive_upsampling:
h0 += self._additive_upsampling(inputs=inputs_nonskip)
r1 = h0 + inputs_skip
r2 = self._conv3d_block(inputs=h0, training=training)
h1 = self._residual_block(inputs=[r2, r1], training=training)
return h1
def test_check_inputs():
"""
Test check_inputs by confirming that it accepts proper
types and handles a few simple cases.
"""
# Check inputs list - Pass
assert layer_util.check_inputs([], 0) is None
# Check inputs tuple - Pass
assert layer_util.check_inputs((), 0) is None
# Check inputs int - Fail
with pytest.raises(ValueError) as execinfo:
layer_util.check_inputs(0, 0)
msg = " ".join(execinfo.value.args[0].split())
assert "Inputs should be a list or tuple" in msg
# Check inputs float - Fail
with pytest.raises(ValueError) as execinfo:
layer_util.check_inputs(0.0, 0)
msg = " ".join(execinfo.value.args[0].split())
assert "Inputs should be a list or tuple" in msg
# Check size float - Fail
with pytest.raises(ValueError) as execinfo:
layer_util.check_inputs([1], 0.5)
msg = " ".join(execinfo.value.args[0].split())
assert "Inputs should be a list or tuple of size" in msg
# Check size 0 - Pass
assert layer_util.check_inputs([], 0) is None
assert layer_util.check_inputs((), 0) is None
# Check size 0 - Fail
with pytest.raises(ValueError) as execinfo:
layer_util.check_inputs([0], 0)
msg = " ".join(execinfo.value.args[0].split())
assert "Inputs should be a list or tuple of size" in msg
with pytest.raises(ValueError) as execinfo:
layer_util.check_inputs((0, ), 0)
msg = " ".join(execinfo.value.args[0].split())
assert "Inputs should be a list or tuple of size" in msg
# Check size 1 - Pass
assert layer_util.check_inputs([0], 1) is None
assert layer_util.check_inputs((0, ), 1) is None
# Check size 1 - Fail
with pytest.raises(ValueError) as execinfo:
layer_util.check_inputs([], 1)
msg = " ".join(execinfo.value.args[0].split())
assert "Inputs should be a list or tuple of size" in msg
with pytest.raises(ValueError) as execinfo:
layer_util.check_inputs((), 1)
msg = " ".join(execinfo.value.args[0].split())
assert "Inputs should be a list or tuple of size" in msg