TensorFlow实战——CNN（Inception-v3）

本文：http://blog.csdn.net/u011239443/article/details/73008131

我们之前介绍的神经网络层与层之间都是一对一的，而Inception-v3模型存在一对多和多对一。

• 一对多：用不同或者相同的过滤器过滤出多个层
• 多对一：将多层合并，即深度会相加。

tensorflow github 中有Inception-v3模型的实现：https://github.com/tensorflow/tensorflow/blob/fc1567c78b3746b44aa50373489a767afbb95d2b/tensorflow/contrib/slim/python/slim/nets/inception_v3.py

由于大部分代码雷同，所以我们就讲解最后一个Mixed的代码：

首先我们来介绍下一个可以来帮我们创建卷积神经网络层的工具：layers.conv2d。它可以很简洁的帮我们创建卷积神经网络卷积层，它本质是convolution函数，我们可以看下convolution函数：

def convolution(inputs,
                num_outputs,
                kernel_size,
                stride=1,
                padding='SAME',
                data_format=None,
                rate=1,
                activation_fn=nn.relu,
                normalizer_fn=None,
                normalizer_params=None,
                weights_initializer=initializers.xavier_initializer(),
                weights_regularizer=None,
                biases_initializer=init_ops.zeros_initializer(),
                biases_regularizer=None,
                reuse=None,
                variables_collections=None,
                outputs_collections=None,
                trainable=True,
                scope=None):

  """
  Args:
    inputs: a Tensor of rank N+2 of shape
      `[batch_size] + input_spatial_shape + [in_channels]` if data_format does
      not start with "NC" (default), or
      `[batch_size, in_channels] + input_spatial_shape` if data_format starts
      with "NC".
    num_outputs: integer, the number of output filters.
    kernel_size: a sequence of N positive integers specifying the spatial
      dimensions of of the filters.  Can be a single integer to specify the same
      value for all spatial dimensions.
    stride: a sequence of N positive integers specifying the stride at which to
     ......
    scope: Optional scope for `variable_scope`.
'''

我们可以看到参数：

• inputs：输入数据
• num_outputs：输出深度
• kernel_size：窗口尺寸
• stride=1：步长，默认为1
• padding=‘SAME’：是否填充0，默认填充

接下来我们来看下这部分的代码：

• Branch_0：

从net，用深度为320，1×1的过滤器，过滤出branch_0。

      end_point = 'Mixed_7b'
      with variable_scope.variable_scope(end_point):
        with variable_scope.variable_scope('Branch_0'):
          branch_0 = layers.conv2d(
              net, depth(320), [1, 1], scope='Conv2d_0a_1x1')

• Branch_1：

先从net，用深度为384，1×1的过滤器，过滤出branch_1。
再从branch_1，分别用一个深度为384，1×3的过滤器和深度为384，3×1的过滤器，过滤出两层。
最后，将这两层合并，赋值给branch_1，concat中参数3，代表在是在第三个维度，即深度上，对两者进行合并。

        with variable_scope.variable_scope('Branch_1'):
          branch_1 = layers.conv2d(
              net, depth(384), [1, 1], scope='Conv2d_0a_1x1')
          branch_1 = array_ops.concat(
              [
                  layers.conv2d(
                      branch_1, depth(384), [1, 3], scope='Conv2d_0b_1x3'),
                  layers.conv2d(
                      branch_1, depth(384), [3, 1], scope='Conv2d_0b_3x1')
              ],
              3)

• Branch_2 & Branch_3

前面两个操作类似，可以参考图片自己理解代码：

        with variable_scope.variable_scope('Branch_2'):
          branch_2 = layers.conv2d(
              net, depth(448), [1, 1], scope='Conv2d_0a_1x1')
          branch_2 = layers.conv2d(
              branch_2, depth(384), [3, 3], scope='Conv2d_0b_3x3')
          branch_2 = array_ops.concat(
              [
                  layers.conv2d(
                      branch_2, depth(384), [1, 3], scope='Conv2d_0c_1x3'),
                  layers.conv2d(
                      branch_2, depth(384), [3, 1], scope='Conv2d_0d_3x1')
              ],
              3)
        with variable_scope.variable_scope('Branch_3'):
          branch_3 = layers_lib.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')
          branch_3 = layers.conv2d(
              branch_3, depth(192), [1, 1], scope='Conv2d_0b_1x1')

最后，将branch_0, branch_1, branch_2, branch_3进行合并：

        net = array_ops.concat([branch_0, branch_1, branch_2, branch_3], 3)
      end_points[end_point] = net
      if end_point == final_endpoint:
return net, end_points