How to use tf.nn.fused_batch_norm

I need simple examples with which I can learn how to use tf.nn.fused_batch_norm

. (I couldn't find any of the google searches.)

In particular, I want to understand exactly the input parts mean=None, variance=None

to a function. In the inference step , I calculate the mean and variance of the population, say tf.nn.moments, and then put them into a function with these input arguments?

And I use this feature just like any other activations like ReLU etc.