def compile_time_operation(self, learning_option, cluster):
"""
define mean-variance normalization(MVN) operation for input tensor.
"""
# get input
input_ = self.get_input('input')
indim = self.get_dimension('input')
# get attr
# optional field
normalize_variance = self.get_attr('normalize_variance', default=True)
across_channels = self.get_attr('across_channels', default=False)
eps = float(self.get_attr('epsilon', default=10**-9))
mvn = L.MVN(input_,
name=self.name,
normalize_variance=normalize_variance,
across_channels=across_channels,
epsilon=eps)
#set output dimension
outdim = indim
self.set_output('output', mvn)
self.set_dimension('output', outdim)
def test_mvn2(self):
n = caffe.NetSpec()
n.input1 = L.Input(shape=make_shape([6, 4, 64, 64]))
n.bnll1 = L.MVN(n.input1,
normalize_variance=False,
across_channels=True,
eps=0.01)
self._test_model(*self._netspec_to_model(n, 'mvn2'))
def test_mvn(self):
n = caffe.NetSpec()
n.input1 = L.Input(shape=make_shape([6, 4, 64, 64]))
n.bnll1 = L.MVN(n.input1, eps=0.01)
self._test_model(*self._netspec_to_model(n, 'mvn'))
def MVN(self, bottom, name='mvn'): return L.MVN(bottom, name=name)