tf.nn.fused_batch_norm(
x,
scale,
offset,
mean=None,
variance=None,
epsilon=0.001,
data_format='NHWC',
is_training=True,
name=None
)
Defined in tensorflow/python/ops/nn_impl.py
.
Batch normalization.
As described in http://arxiv.org/abs/1502.03167.
Args:
x
: InputTensor
of 4 dimensions.scale
: ATensor
of 1 dimension for scaling.offset
: ATensor
of 1 dimension for bias.mean
: ATensor
of 1 dimension for population mean used for inference.variance
: ATensor
of 1 dimension for population variance used for inference.epsilon
: A small float number added to the variance of x.data_format
: The data format for x. Either "NHWC" (default) or "NCHW".is_training
: A bool value to specify if the operation is used for training or inference.name
: A name for this operation (optional).
Returns:
y
: A 4D Tensor for the normalized, scaled, offsetted x.batch_mean
: A 1D Tensor for the mean of x.batch_var
: A 1D Tensor for the variance of x.
Raises:
ValueError
: If mean or variance is not None when is_training is True.