MobileNet V1 网络结构的原理与 Tensorflow2.0 实现

MobileNet 网络结构

介绍 MobileNet 之前，我们需要先了解一下它的重要组件——深度级可分离卷积（Depthwise separable convolution）网络，这种网络可以被进一步分解成两个部分：对输入的每个通道使用单通道卷积核进行卷积（depthwise convolution）、使用 1x1 卷积核将输入通道数转变成输出通道数（pointwise convolution）。举例如下图所示：

在第一部分中，深度级卷积和普通卷积不同，对于标准卷积其卷积核是用在所有的输入通道上的，而深度级可分离卷积针对每个输入通道采用不同的卷积核，即一个卷积核对应一个输入通道，所以我们说这个操作是发生在深度级上的。

在第二部分中，pointwise convolution 其实就和普通卷积操作相同，只不过使用的卷积核尺寸是 1x1 的。

深度级可分离卷积和一个普通卷积的整体效果差不多，但前者会大大减少计算量和模型参数量，具体可参考卷积、可分离卷积的参数量与计算量的对比。

在 MobileNet 网络结构中，有 13 个深度级可分离卷积网络被使用。

代码实现

用函数实现

import tensorflow as tf

def conv_block(inputs, filters, kernel_size = (3,3), strides=(1,1)):
    x = tf.keras.layers.Conv2D(filters=filters, kernel_size=kernel_size, strides=strides,     
                               padding='SAME', use_bias=False)(inputs)
    x = tf.keras.layers.BatchNormalization()(x)
    out = tf.keras.layers.Activation('relu')(x)
 
    return out

def depthwise_conv_block(inputs,
                         pointwise_conv_filters,
                         strides=(1,1)):
    x = tf.keras.layers.DepthwiseConv2D(kernel_size=(3,3), strides=strides, padding='SAME',
                                        use_bias=False)(inputs)
    x = tf.keras.layers.BatchNormalization()(x)
    x = tf.keras.layers.Activation('relu')(x)
 
    x = tf.keras.layers.Conv2D(filters=pointwise_conv_filters, kernel_size=(1,1), 
                               padding='SAME', use_bias=False)(x)
    x = tf.keras.layers.BatchNormalization()(x)
    out = tf.keras.layers.Activation('relu')(x)
    return out

def mobilenet_v1(inputs,
                 classes):
    # [32, 32, 3] => [16, 16, 32]
    x = conv_block(inputs, 32, strides=(2,2))
    # [16, 16, 32] => [16, 16, 64]
    x = depthwise_conv_block(x, 64)
    # [16, 16, 64] => [8, 8, 128]
    x = depthwise_conv_block(x, 128, strides=(2,2))
    # [8, 8, 128] => [8, 8, 128]
    x = depthwise_conv_block(x, 128)
    # [8, 8, 128] => [4, 4, 256]
    x = depthwise_conv_block(x, 256, strides=(2, 2))
    # [4, 4, 256] => [4, 4, 256]
    x = depthwise_conv_block(x, 256)
    # [4, 4, 256] => [2, 2, 512]
    x = depthwise_conv_block(x, 512, strides=(2, 2))
    # [2, 2, 512] => [2, 2, 512]
    x = depthwise_conv_block(x, 512)
    # [2, 2, 512] => [2, 2, 512]
    x = depthwise_conv_block(x, 512)
    # [2, 2, 512] => [2, 2, 512]
    x = depthwise_conv_block(x, 512)
    # [2, 2, 512] => [2, 2, 512]
    x = depthwise_conv_block(x, 512)
    # [2, 2, 512] => [2, 2, 512]
    x = depthwise_conv_block(x, 512)
    # [2, 2, 512] => [1, 1, 1024]
    x = depthwise_conv_block(x, 1024, strides=(2,2))
    # [1, 1, 1024] => [1, 1, 1024]
    x = depthwise_conv_block(x, 1024)
    
    # [1, 1, 1024] => (1024,)
    x = tf.keras.layers.GlobalAveragePooling2D()(x)
    # (1024,) => (classes,)
    pred = tf.keras.layers.Dense(classes, activation='softmax')(x)
 
    return pred

inputs = np.zeros((10, 32, 32, 3), dtype=np.float32)
classes = 10
output = mobilenet_v1(inputs, classes)

output.shape

TensorShape([10, 10])

用类实现

import tensorflow as tf

class conv_block(tf.keras.Model):
    def __init__(self, filters, kernel_size = (3,3), strides=(1,1)):
        super().__init__()
        self.listLayers = []
        self.listLayers.append(tf.keras.layers.Conv2D(filters=filters, kernel_size=kernel_size, strides=strides,     
                              padding='SAME', use_bias=False))
        self.listLayers.append(tf.keras.layers.BatchNormalization())
        self.listLayers.append(tf.keras.layers.Activation('relu'))
        
    def call(self, x):
        for layer in self.listLayers.layers:
            x = layer(x)
        return x

class depthwise_conv_block(tf.keras.Model):
    def __init__(self, pointwise_conv_filters, strides=(1,1)):
        super().__init__()
        self.listLayers = []
        self.listLayers.append(tf.keras.layers.DepthwiseConv2D(kernel_size=(3,3), strides=strides, padding='SAME',
                                        use_bias=False))
        self.listLayers.append(tf.keras.layers.BatchNormalization())
        self.listLayers.append(tf.keras.layers.Activation('relu'))
        self.listLayers.append(tf.keras.layers.Conv2D(filters=pointwise_conv_filters, kernel_size=(1,1), 
                               padding='SAME', use_bias=False))
        self.listLayers.append(tf.keras.layers.BatchNormalization())
        self.listLayers.append(tf.keras.layers.Activation('relu'))
        
    def call(self, x):
        for layer in self.listLayers.layers:
            x = layer(x)
        return x

class mobilenet_v1(tf.keras.Model):
    def __init__(self, classes):
        super().__init__()
        self.listLayers = []
        self.listLayers.append(conv_block(32, strides=(2,2)))
        self.listLayers.append(depthwise_conv_block(64))
        self.listLayers.append(depthwise_conv_block(128, strides=(2,2)))
        self.listLayers.append(depthwise_conv_block(128))
        self.listLayers.append(depthwise_conv_block(256, strides=(2, 2)))
        self.listLayers.append(depthwise_conv_block(256))
        self.listLayers.append(depthwise_conv_block(512, strides=(2, 2)))
        self.listLayers.append(depthwise_conv_block(512))
        self.listLayers.append(depthwise_conv_block(512))
        self.listLayers.append(depthwise_conv_block(512))
        self.listLayers.append(depthwise_conv_block(512))
        self.listLayers.append(depthwise_conv_block(512))
        self.listLayers.append(depthwise_conv_block(1024, strides=(2,2)))
        self.listLayers.append(depthwise_conv_block(1024))
        self.listLayers.append(tf.keras.layers.GlobalAveragePooling2D())
        self.listLayers.append(tf.keras.layers.Dense(classes, activation='softmax'))
        
    def call(self, x):
        for layer in self.listLayers.layers:
            x = layer(x)
        return x

inputs = np.zeros((10, 32, 32, 3), dtype=np.float32)
model = mobilenet_v1(10)
model(inputs).shape

TensorShape([10, 10])