def extract_batch_norm(
self,
input_name,
output_name,
scope_id,
data_format="NCHW",
axis=1,
eps=1e-3,
layer_names=["bn", "moving_mean", "moving_variance"]):
assert (data_format == "NCHW")
mean, var = self.get_weights(scope_id, layer_names)
layer = caffe_net.LayerParameter(name=output_name,
type='BatchNorm',
bottom=[input_name],
top=[output_name])
layer.batch_norm_param(axis=axis, eps=eps)
layer.add_data(mean, var)
self.caffe_model.add_layer(layer)
if (self.data_dict[input_name] is not None):
input_data, input_shape, inv_transpose_dims = self.preprocess_nchwc8_nchw_input(
input_name, axis)
output_data = Operators.batch_norm(input_data, mean, var, eps,
output_name)
self.data_dict[output_name] = self.postprocess_nchwc8_nchw_output(
output_data, input_shape, inv_transpose_dims)
else:
self.data_dict[output_name] = None
gamma = self.get_weights(scope_id, [layer_names[0], "gamma"])
if (gamma is not None):
scale_name = self.extract_scale(output_name, output_name + "_s",
scope_id, data_format, axis,
[layer_names[0], "gamma", "beta"])
self.data_dict[output_name] = self.data_dict[scale_name]
output_name = scale_name
return output_name
