slim.arg_scope嵌套理解（超简单）

def new_arg_sc():

    with slim.arg_scope([slim.conv2d, slim.fully_connected],

              trainable = True,

              activation_fn = tf.nn.relu,

              weights_initializer = tf.truncated_normal_initializer(stddev=0.01),

              weights_regularizer = slim.l2_regularizer(0.0001)):

        with slim.arg_scope([slim.conv2d],

                  kernel_size = [3, 3],

                  padding = 'SAME',

                  normalizer_fn = slim.batch_norm) as sc:

            return sc

嵌套的第一层 slim.arg_scope 代表设置默认的 slim.conv2d 和 slim.fully_connected 的通用参数

第二层 slim.arg_scope则在原来设置的默认参数的基础上设置 slim.conv2d 其他三个 slim.fully_connected 没有的参数，即 kernel_size、padding、normalizer_fn。

---

返回的 sc 可以作为其他slim.arg_scope的参数进行默认参数的传递：

如：with slim.arg_scope(sc):