【深度学习计算机视觉】Tensorflow VGG源码深入讲解

本文分享内容来自《深度学习计算机视觉实战》一书​：

VGG 是由牛津大学 Visual Geometry Group 组在 2014 年提出的，算法的名字也由此而来。

VGG 的提出证明了网络深度的增加可以提升网络的性能。VGG 常用的网络结构有两种：VGG-16和 VGG-19，网络结构在本质上没有差异，只是网络的深度不一样。VGG 相比以前的网络有一个重要的改进，使用连续的小的卷积核代替较大的卷积核，如使用两个 3×3 的卷积核代替一个 5×5 的卷积核。这种替换不仅可以增加网络的深度，还可以显著地减少网络的计算量，对性能的提升有较大作用。

VGG 的网络结构如图 1.12 所示。图 1.12 中 D 列所示为 VGG-16，该网络包含 16 个隐藏层，其中 13 个卷积层和 3 个全连接层；E 列所示为 VGG-19，该网络包含 19 个隐藏层，其中 16 个卷的卷积都是采用 3×3 的卷积核，池化使用的是 kernel 大小为 2×2的最大池化操作。

图 1.12
VGG 网络是很多算法中特征提取的主干网络（backbone），具有重要的意义，但是 VGG 的显著缺点是参数量很大，需要耗费较大的计算资源，参数量很大部分来源于 3 个全连接层，所以将VGG 作为特征提取网络时，常用的做法是只使用网络的卷积部分。

VGG 网络的深入研究可以参考论文 Very Deep Convolutional Networks for Large-Scale Image Recognition。
VGG16的代码如下：

import inspect
import os

import numpy as np
import tensorflow as tf
import time

# 用于计算图像平均减法，以便数据标准化
VGG_MEAN = [103.939, 116.779, 123.68]

# VGG16
class Vgg16:
    def __init__(self, vgg16_npy_path=None):
        if vgg16_npy_path is None:
            path = inspect.getfile(Vgg16)
            path = os.path.abspath(os.path.join(path, os.pardir))
            path = os.path.join(path, "vgg16.npy")
            vgg16_npy_path = path
            print(path)

        self.data_dict = np.load(vgg16_npy_path, encoding='latin1').item()
        print("npy file loaded")

    # 网络搭建
    def build(self, rgb):
        """
        load variable from npy to build the VGG

        :param rgb: rgb image [batch, height, width, 3] values scaled [0, 1]
        """
        # 计算运行时间
        start_time = time.time()
        print("build model started")
        rgb_scaled = rgb * 255.0

        # 图像通道转换， RGB to BGR
        red, green, blue = tf.split(axis=3, num_or_size_splits=3, value=rgb_scaled)
        assert red.get_shape().as_list()[1:] == [224, 224, 1]
        assert green.get_shape().as_list()[1:] == [224, 224, 1]
        assert blue.get_shape().as_list()[1:] == [224, 224, 1]
        bgr = tf.concat(axis=3, values=[
            blue - VGG_MEAN[0],
            green - VGG_MEAN[1],
            red - VGG_MEAN[2],
        ])
        assert bgr.get_shape().as_list()[1:] == [224, 224, 3]
        # 搭建网络层，可以自己对照VGG16的网络结构学习
        self.conv1_1 = self.conv_layer(bgr, "conv1_1")
        self.conv1_2 = self.conv_layer(self.conv1_1, "conv1_2")
        self.pool1 = self.max_pool(self.conv1_2, 'pool1')

        self.conv2_1 = self.conv_layer(self.pool1, "conv2_1")
        self.conv2_2 = self.conv_layer(self.conv2_1, "conv2_2")
        self.pool2 = self.max_pool(self.conv2_2, 'pool2')

        self.conv3_1 = self.conv_layer(self.pool2, "conv3_1")
        self.conv3_2 = self.conv_layer(self.conv3_1, "conv3_2")
        self.conv3_3 = self.conv_layer(self.conv3_2, "conv3_3")
        self.pool3 = self.max_pool(self.conv3_3, 'pool3')

        self.conv4_1 = self.conv_layer(self.pool3, "conv4_1")
        self.conv4_2 = self.conv_layer(self.conv4_1, "conv4_2")
        self.conv4_3 = self.conv_layer(self.conv4_2, "conv4_3")
        self.pool4 = self.max_pool(self.conv4_3, 'pool4')

        self.conv5_1 = self.conv_layer(self.pool4, "conv5_1")
        self.conv5_2 = self.conv_layer(self.conv5_1, "conv5_2")
        self.conv5_3 = self.conv_layer(self.conv5_2, "conv5_3")
        self.pool5 = self.max_pool(self.conv5_3, 'pool5')

        self.fc6 = self.fc_layer(self.pool5, "fc6")
        assert self.fc6.get_shape().as_list()[1:] == [4096]
        self.relu6 = tf.nn.relu(self.fc6)

        # 全连接
        self.fc7 = self.fc_layer(self.relu6, "fc7")
        self.relu7 = tf.nn.relu(self.fc7)

        self.fc8 = self.fc_layer(self.relu7, "fc8")

        self.prob = tf.nn.softmax(self.fc8, name="prob")

        self.data_dict = None
        print(("build model finished: %ds" % (time.time() - start_time)))

    # 平均池化
    def avg_pool(self, bottom, name):
        return tf.nn.avg_pool(bottom, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', name=name)

    # 最大池化
    def max_pool(self, bottom, name):
        return tf.nn.max_pool(bottom, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', name=name)

    # 卷积层
    def conv_layer(self, bottom, name):
        with tf.variable_scope(name):
            filt = self.get_conv_filter(name)

            conv = tf.nn.conv2d(bottom, filt, [1, 1, 1, 1], padding='SAME')

            conv_biases = self.get_bias(name)
            bias = tf.nn.bias_add(conv, conv_biases)

            relu = tf.nn.relu(bias)
            return relu
    # 全连接
    def fc_layer(self, bottom, name):
        with tf.variable_scope(name):
            shape = bottom.get_shape().as_list()
            dim = 1
            for d in shape[1:]:
                dim *= d
            x = tf.reshape(bottom, [-1, dim])

            weights = self.get_fc_weight(name)
            biases = self.get_bias(name)

            # Fully connected layer. Note that the '+' operation automatically
            # broadcasts the biases.
            fc = tf.nn.bias_add(tf.matmul(x, weights), biases)

            return fc
    # 获取卷积核
    def get_conv_filter(self, name):
        return tf.constant(self.data_dict[name][0], name="filter")
    # 获取偏置
    def get_bias(self, name):
        return tf.constant(self.data_dict[name][1], name="biases")
    # 获取权重
    def get_fc_weight(self, name):
        return tf.constant(self.data_dict[name][0], name="weights")

VGG19的代码如下：

import os
import tensorflow as tf

import numpy as np
import time
import inspect

VGG_MEAN = [103.939, 116.779, 123.68]

# VGG19
class Vgg19:
    def __init__(self, vgg19_npy_path=None):
        if vgg19_npy_path is None:
            path = inspect.getfile(Vgg19)
            path = os.path.abspath(os.path.join(path, os.pardir))
            path = os.path.join(path, "vgg19.npy")
            vgg19_npy_path = path
            print(vgg19_npy_path)

        self.data_dict = np.load(vgg19_npy_path, encoding='latin1').item()
        print("npy file loaded")

    # 网络搭建
    def build(self, rgb):
        """
        load variable from npy to build the VGG

        :param rgb: rgb image [batch, height, width, 3] values scaled [0, 1]
        """

        start_time = time.time()
        print("build model started")
        rgb_scaled = rgb * 255.0

        # Convert RGB to BGR
        red, green, blue = tf.split(axis=3, num_or_size_splits=3, value=rgb_scaled)
        assert red.get_shape().as_list()[1:] == [224, 224, 1]
        assert green.get_shape().as_list()[1:] == [224, 224, 1]
        assert blue.get_shape().as_list()[1:] == [224, 224, 1]
        bgr = tf.concat(axis=3, values=[
            blue - VGG_MEAN[0],
            green - VGG_MEAN[1],
            red - VGG_MEAN[2],
        ])
        assert bgr.get_shape().as_list()[1:] == [224, 224, 3]

        self.conv1_1 = self.conv_layer(bgr, "conv1_1")
        self.conv1_2 = self.conv_layer(self.conv1_1, "conv1_2")
        self.pool1 = self.max_pool(self.conv1_2, 'pool1')

        self.conv2_1 = self.conv_layer(self.pool1, "conv2_1")
        self.conv2_2 = self.conv_layer(self.conv2_1, "conv2_2")
        self.pool2 = self.max_pool(self.conv2_2, 'pool2')

        # VGG19和VGG16的网络层差异主要在后面部分，多了三层卷积
        self.conv3_1 = self.conv_layer(self.pool2, "conv3_1")
        self.conv3_2 = self.conv_layer(self.conv3_1, "conv3_2")
        self.conv3_3 = self.conv_layer(self.conv3_2, "conv3_3")
        self.conv3_4 = self.conv_layer(self.conv3_3, "conv3_4")
        self.pool3 = self.max_pool(self.conv3_4, 'pool3')

        self.conv4_1 = self.conv_layer(self.pool3, "conv4_1")
        self.conv4_2 = self.conv_layer(self.conv4_1, "conv4_2")
        self.conv4_3 = self.conv_layer(self.conv4_2, "conv4_3")
        self.conv4_4 = self.conv_layer(self.conv4_3, "conv4_4")
        self.pool4 = self.max_pool(self.conv4_4, 'pool4')

        self.conv5_1 = self.conv_layer(self.pool4, "conv5_1")
        self.conv5_2 = self.conv_layer(self.conv5_1, "conv5_2")
        self.conv5_3 = self.conv_layer(self.conv5_2, "conv5_3")
        self.conv5_4 = self.conv_layer(self.conv5_3, "conv5_4")
        self.pool5 = self.max_pool(self.conv5_4, 'pool5')

        self.fc6 = self.fc_layer(self.pool5, "fc6")
        assert self.fc6.get_shape().as_list()[1:] == [4096]
        self.relu6 = tf.nn.relu(self.fc6)

        self.fc7 = self.fc_layer(self.relu6, "fc7")
        self.relu7 = tf.nn.relu(self.fc7)

        self.fc8 = self.fc_layer(self.relu7, "fc8")

        self.prob = tf.nn.softmax(self.fc8, name="prob")

        self.data_dict = None
        print(("build model finished: %ds" % (time.time() - start_time)))

    # 下述函数含义与VGG16相同
    def avg_pool(self, bottom, name):
        return tf.nn.avg_pool(bottom, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', name=name)

    def max_pool(self, bottom, name):
        return tf.nn.max_pool(bottom, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', name=name)

    def conv_layer(self, bottom, name):
        with tf.variable_scope(name):
            filt = self.get_conv_filter(name)

            conv = tf.nn.conv2d(bottom, filt, [1, 1, 1, 1], padding='SAME')

            conv_biases = self.get_bias(name)
            bias = tf.nn.bias_add(conv, conv_biases)

            relu = tf.nn.relu(bias)
            return relu

    def fc_layer(self, bottom, name):
        with tf.variable_scope(name):
            shape = bottom.get_shape().as_list()
            dim = 1
            for d in shape[1:]:
                dim *= d
            x = tf.reshape(bottom, [-1, dim])

            weights = self.get_fc_weight(name)
            biases = self.get_bias(name)

            # Fully connected layer. Note that the '+' operation automatically
            # broadcasts the biases.
            fc = tf.nn.bias_add(tf.matmul(x, weights), biases)

            return fc

    def get_conv_filter(self, name):
        return tf.constant(self.data_dict[name][0], name="filter")

    def get_bias(self, name):
        return tf.constant(self.data_dict[name][1], name="biases")

    def get_fc_weight(self, name):
        return tf.constant(self.data_dict[name][0], name="weights")