def test_batchnormalization(self):
K.set_image_dim_ordering("th")
input_data = np.random.random_sample([2, 5, 32, 32])
zlayer = ZLayer.BatchNormalization(axis=1, input_shape=(5, 32, 32))
klayer = KLayer.BatchNormalization(axis=1, input_shape=(5, 32, 32))
self.compare_layer(klayer, zlayer, input_data,
WeightsConverter.convert_batchnormalization)
K.set_image_dim_ordering("tf")
input_data2 = np.random.random_sample([2, 32, 32, 4])
zlayer = ZLayer.BatchNormalization(axis=-1,
dim_ordering="tf",
input_shape=(32, 32, 4))
klayer = KLayer.BatchNormalization(axis=-1, input_shape=(32, 32, 4))
self.compare_layer(klayer, zlayer, input_data2,
WeightsConverter.convert_batchnormalization)
def _to_tensor(self):
input = self.model_inputs[0]
rank = len(input.zvalue.shape)
if (rank == 4):
epsilon = float(self.onnx_attr['epsilon']
) if "epsilon" in self.onnx_attr else 0.001
momentum = float(self.onnx_attr['momentum'] if "momentum" in
self.onnx_attr else 0.99)
dim_ordering = "th"
if len(self._input_list) == 5:
mean = self._input_list[3].zvalue
variance = self._input_list[4].zvalue
else:
mean = self._input_list[1].zvalue
variance = self._input_list[2].zvalue
batch_norm = zlayers.BatchNormalization(epsilon=epsilon,
momentum=momentum,
dim_ordering=dim_ordering)
norm_tensor = batch_norm(input.zvalue)
norm_tensor.node.element().set_running_mean(mean)
norm_tensor.node.element().set_running_std(variance)
return norm_tensor
else:
raise Exception("not supported.")