refinet网络将卷积层中的conv6-1以及conv6-2等相关联层删除，然后对conv4-3以及coonv5-3卷积层进行bn归一化处理代码

这里是一个使用TensorFlow实现的refinet网络的代码示例，包括了对conv4-3和conv5-3卷积层的BN归一化处理。

import tensorflow as tf

def conv_bn_relu(inputs, filters, kernel_size, strides, padding, training):
    x = tf.layers.conv2d(inputs=inputs, filters=filters, kernel_size=kernel_size, strides=strides, padding=padding)
    x = tf.layers.batch_normalization(x, training=training)
    x = tf.nn.relu(x)
    return x

def refinet(inputs, training):
    # Conv1
    x = conv_bn_relu(inputs, filters=64, kernel_size=3, strides=1, padding='same', training=training)
    x = conv_bn_relu(x, filters=64, kernel_size=3, strides=1, padding='same', training=training)
    x = tf.layers.max_pooling2d(x, pool_size=2, strides=2, padding='same')

    # Conv2
    x = conv_bn_relu(x, filters=128, kernel_size=3, strides=1, padding='same', training=training)
    x = conv_bn_relu(x, filters=128, kernel_size=3, strides=1, padding='same', training=training)
    x = tf.layers.max_pooling2d(x, pool_size=2, strides=2, padding='same')

    # Conv3
    x = conv_bn_relu(x, filters=256, kernel_size=3, strides=1, padding='same', training=training)
    x = conv_bn_relu(x, filters=256, kernel_size=3, strides=1, padding='same', training=training)
    x = conv_bn_relu(x, filters=256, kernel_size=3, strides=1, padding='same', training=training)
    x = tf.layers.max_pooling2d(x, pool_size=2, strides=2, padding='same')

    # Conv4
    x = conv_bn_relu(x, filters=512, kernel_size=3, strides=1, padding='same', training=training)
    x = conv_bn_relu(x, filters=512, kernel_size=3, strides=1, padding='same', training=training)
    conv4_3 = conv_bn_relu(x, filters=512, kernel_size=3, strides=1, padding='same', training=training)

    # Conv5
    x = tf.layers.max_pooling2d(conv4_3, pool_size=2, strides=2, padding='same')
    x = conv_bn_relu(x, filters=512, kernel_size=3, strides=1, padding='same', training=training)
    x = conv_bn_relu(x, filters=512, kernel_size=3, strides=1, padding='same', training=training)
    conv5_3 = conv_bn_relu(x, filters=512, kernel_size=3, strides=1, padding='same', training=training)

    # Conv6
    x = tf.layers.max_pooling2d(conv5_3, pool_size=2, strides=2, padding='same')
    x = conv_bn_relu(x, filters=1024, kernel_size=3, strides=1, padding='same', training=training)
    x = conv_bn_relu(x, filters=1024, kernel_size=3, strides=1, padding='same', training=training)
    x = conv_bn_relu(x, filters=1024, kernel_size=3, strides=1, padding='same', training=training)

    # Conv7
    conv6_1 = tf.layers.conv2d(inputs=x, filters=256, kernel_size=1, strides=1, padding='same')
    conv6_1_bn = tf.layers.batch_normalization(conv6_1, training=training)
    conv6_1_relu = tf.nn.relu(conv6_1_bn)

    conv6_2 = tf.layers.conv2d(inputs=x, filters=512, kernel_size=3, strides=2, padding='same')
    conv6_2_bn = tf.layers.batch_normalization(conv6_2, training=training)
    conv6_2_relu = tf.nn.relu(conv6_2_bn)

    return conv4_3, conv5_3, conv6_1_relu, conv6_2_relu

在上面的代码中，`conv_bn_relu`函数是一个便捷的函数，用于定义一个卷积层、Batch Normalization层和ReLU激活函数的组合。`refinet`函数是整个RefineNet的主要实现，其中包括了对Conv4-3和Conv5-3卷积层的BN归一化处理。