BN,GN,IN,LN的TensorFlow代码实现

#BN
def BN(input, bn_training, name):
    # assert isinstance(is_training, (ops.Tensor, variables.Variable)) and is_training.dtype == tf.bool
    def bn(input, bn_training, name, reuse=None):
        with tf.variable_scope('bn', reuse=reuse):
            output = tf.layers.batch_normalization(input, training=bn_training, name=name)
        return output
    return tf.cond(
        bn_training,
        lambda: bn(input, bn_training=True, name=name, reuse=None),
        lambda: bn(input, bn_training=False, name=name, reuse=True),
    )
#LN
def Layernorm(x, gamma, beta):
    # x_shape:[B, H, W, C]
    results = 0.
    eps = 1e-5
    x = tf.transpose(x, [0, 3, 1, 2])  # [B,C,H,W]
    x_mean = np.mean(x, axis=(1, 2, 3), keepdims=True)
    x_var = np.var(x, axis=(1, 2, 3), keepdims=True0)
    x_normalized = (x - x_mean) / np.sqrt(x_var + eps)
    results = gamma * x_normalized + beta
    results = tf.transpose(results, [0, 2, 3, 1])
    return results

#IN
def Instancenorm(x, gam