resnet50 结构详解 resnet50代码详解

Resnet50的代码不是由笔者编写，笔者只对代码进行讲解，方便后续使用。原作者博客链接。 为了节省篇幅这里不贴出代码，请访问原作者GitHub查看代码。

在阅读本博客前请先了解残差网络的结构和原理，推荐博客。

1.ResNet50的网络结构

Resnet50包含两个基本的模块：Conv Block和Identity Block。这两个模块的结构图如下所示：

从图中可以看到，Identity Block的输出和输入必须是有相同的形状（不然残差边和input不能相加），这导致卷积提取到的特征长、宽和维度不能发生变化，而Conv Block在残差边加入了卷积操作，可以对输入矩阵的形状进行调整，使得残差边和卷积可以进行相加。

ResNet50的完整结构图如下图所示：

2. ResNet50

def ResNet50(input_shape=[224,224,3],classes=1000):
    # [224,224,3]
    img_input = Input(shape=input_shape)
    x = ZeroPadding2D((3, 3))(img_input)   # [230,230,3]
    # [112,112,64]
    x = Conv2D(64, (7, 7), strides=(2, 2), name='conv1')(x)   #[112,112,64]
    x = BatchNormalization(name='bn_conv1')(x)
    x = Activation('relu')(x)

    # [56,56,64]
    x = MaxPooling2D((3, 3), strides=(2, 2))(x)

    # [56,56,256]
    x = conv_block(x, 3, [64, 64, 256], stage=2, block='a', strides=(1, 1))
    x = identity_block(x, 3, [64, 64, 256], stage=2, block='b')
    x = identity_block(x, 3, [64, 64, 256], stage=2, block='c')

    # [28,28,512]
    x = conv_block(x, 3, [128, 128, 512], stage=3, block='a')
    x = identity_block(x, 3, [128, 128, 512], stage=3, block='b')
    x = identity_block(x, 3, [128, 128, 512], stage=3, block='c')
    x = identity_block(x, 3, [128, 128, 512], stage=3, block='d')

    # [14,14,1024]
    x = conv_block(x, 3, [256, 256, 1024], stage=4, block='a')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='b')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='c')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='d')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='e')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='f')

    # [7,7,2048]
    x = conv_block(x, 3, [512, 512, 2048], stage=5, block='a')
    x = identity_block(x, 3, [512, 512, 2048], stage=5, block='b')
    x = identity_block(x, 3, [512, 512, 2048], stage=5, block='c')

    # 代替全连接层
    x = AveragePooling2D((7, 7), name='avg_pool')(x)

    # 进行预测
    x = Flatten()(x)
    x = Dense(classes, activation='softmax', name='fc1000')(x)

    model = Model(img_input, x, name='resnet50')

    return model

x = ZeroPadding2D((3, 3))(img_input)：在img_input的外围进行padding=0的填充，填充数量为3个像素点，最后得到的维度为224+3*2 = 230，所以x的维度为230，230，3。

输入图像的尺寸为[224, 224, 3]，经过简单的卷积和池化操作后形状变为[56, 56, 64]。接下来，就是ResNet50网络的重点，Conv Block和Identity Block。下面我们分别以第一次使用Conv Block和Identity Block为例，讲解这两个模块内部是如何操作的。

关于卷积与池化操作时的维度变化可以参考我之前的博客，在这里不做过多赘述，维度变化信息我写在了代码里。

3.Conv Block

Conv Block模块代码如下：

def conv_block(input_tensor, kernel_size, filters, stage, block, strides=(2, 2)):

    # 64,64,256
    filters1, filters2, filters3 = filters

    conv_name_base = 'res' + str(stage) + block + '_branch'
    bn_name_base = 'bn' + str(stage) + block + '_branch'

    # 降维
    x = Conv2D(filters1, (1, 1), strides=strides,
               name=conv_name_base + '2a')(input_tensor)
    x = BatchNormalization(name=bn_name_base + '2a')(x)
    x = Activation('relu')(x)

    # 3x3卷积
    x = Conv2D(filters2, kernel_size, padding='same',
               name=conv_name_base + '2b')(x)
    x = BatchNormalization(name=bn_name_base + '2b')(x)
    x = Activation('relu')(x)

    # 升维
    x = Conv2D(filters3, (1, 1), name=conv_name_base + '2c')(x)
    x = BatchNormalization(name=bn_name_base + '2c')(x)

    # 残差边
    shortcut = Conv2D(filters3, (1, 1), strides=strides,
                      name=conv_name_base + '1')(input_tensor)
    shortcut = BatchNormalization(name=bn_name_base + '1')(shortcut)


    x = layers.add([x, shortcut])
    x = Activation('relu')(x)
    return x

第一次调用Conv Block输入的维度为x.shape = [56, 56, 64]，传入的参数为：
x = conv_block(x, 3, [64, 64, 256], stage=2, block=‘a’, strides=(1, 1))
则Conv Block内：
kernel_szie = 3
filter1 = 64
filter2 = 64
filter3 = 256
strides = (1,1)
stage和block是用来命名确定该层位置的参数。

为了便于查看，以下函数都直接将参数填入

• 命名

conv_name_base = 'res' + '2' + 'a' + '_branch'
bn_name_base = 'bn' + '2' + 'a' + '_branch'

conv_name_base = ‘res2a_branch’
bn_name_base = ‘bn2a_branch’

• 该模块内的第一次卷积操作

# 降维
x = Conv2D(64, (1, 1), strides=(1,1),
               name=conv_name_base + '2a')(input_tensor)
x = BatchNormalization(name=bn_name_base + '2a')(x)
x = Activation('relu')(x)

conv: [56, 56, 64] ==> [56, 56, 64]，name = ‘res2a_branch2a’
BatchNor：形状不发生变化，name = ‘bn2a_branch2a’

• 该模块内的第二次卷积操作

# 3x3卷积
x = Conv2D(64, 3, padding='same',
               name=conv_name_base + '2b')(x)
x = BatchNormalization(name=bn_name_base + '2b')(x)
x = Activation('relu')(x)

conv: [56, 56, 64] ==> [56, 56, 64], name = ‘res2a_branch2b’
BatchNor：形状不发生变化，name = ‘bn2a_branch2b’

• 该模块内的第三次卷积操作

# 升维
x = Conv2D(256, (1, 1), name=conv_name_base + '2c')(x)
x = BatchNormalization(name=bn_name_base + '2c')(x)

conv: [56, 56, 64] ==> [56, 56, 256], name = ‘res2a_branch2c’
BatchNor：形状不发生变化，name = ‘bn2a_branch2c’
输出矩阵的大小最终又回到了[56, 56, 256]

• 残差边

# 残差边
shortcut = Conv2D(256, (1, 1), strides=(1,1),
                      name=conv_name_base + '1')(input_tensor)
shortcut = BatchNormalization(name=bn_name_base + '1')(shortcut)

conv: [56, 56, 64] ==> [56, 56, 256], name = ‘res2a_branch1’
BatchNor: 形状不变, name = ‘bn2a_branch1’
最终残差边的输出[56, 56, 256]和卷积层的输出[56, 56, 256]相加，经激活函数后输出。这一个Conv Block的输入和输出形状没有发生变化。

x = layers.add([x, shortcut])
    x = Activation('relu')(x)
return x

4. Idnetity Block

def identity_block(input_tensor, kernel_size, filters, stage, block):

    filters1, filters2, filters3 = filters

    conv_name_base = 'res' + str(stage) + block + '_branch'
    bn_name_base = 'bn' + str(stage) + block + '_branch'

    # 降维
    x = Conv2D(filters1, (1, 1), name=conv_name_base + '2a')(input_tensor)
    x = BatchNormalization(name=bn_name_base + '2a')(x)
    x = Activation('relu')(x)
    # 3x3卷积
    x = Conv2D(filters2, kernel_size,padding='same', name=conv_name_base + '2b')(x)
    x = BatchNormalization(name=bn_name_base + '2b')(x)
    x = Activation('relu')(x)
    # 升维
    x = Conv2D(filters3, (1, 1), name=conv_name_base + '2c')(x)
    x = BatchNormalization(name=bn_name_base + '2c')(x)

    x = layers.add([x, input_tensor])
    x = Activation('relu')(x)
    return x

第一次输入的维度为x.shape = [56, 56, 256]，传入的参数为：
x = identity_block(x, 3, [64, 64, 256], stage=2, block=‘b’)
则Conv Block内：
kernel_szie = 3
filter1 = 64
filter2 = 64
filter3 = 256
stage和block是用来命名确定该层位置的参数。

为了便于查看，以下函数都直接将参数填入

• 命名

conv_name_base = 'res' + '2'+ 'b'+ '_branch'
bn_name_base = 'bn' + '2' + 'b' + '_branch'

conv_name_base = ‘res2b_branch’
bn_name_base = ‘bn2b_branch’

• 该模块内的第一次卷积

# 降维
x = Conv2D(64, (1, 1), name=conv_name_base + '2a')(input_tensor)
x = BatchNormalization(name=bn_name_base + '2a')(x)
x = Activation('relu')(x)

conv: [56, 56, 256] ==> [56, 56, 64], name = ‘res2a_branch2a’
BatchNor: 形状不变, name = ‘bn2a_branch2a’

• 该模块内的第二次卷积

# 3x3卷积
x = Conv2D(64, kernel_size,padding='same', name=conv_name_base + '2b')(x)
x = BatchNormalization(name=bn_name_base + '2b')(x)
x = Activation('relu')(x)

conv: [56, 56, 64] ==> [56, 56, 64], name = ‘res2a_branch2b’
BatchNor: 形状不变, name = ‘bn2a_branch2b’

• 该模块内的第三次卷积

# 升维
x = Conv2D(256, (1, 1), name=conv_name_base + '2c')(x)
x = BatchNormalization(name=bn_name_base + '2c')(x)

conv: [56, 56, 64] ==> [56, 56, 256], name = ‘res2a_branch2c’
BatchNor: 形状不变, name = ‘bn2a_branch2c’

• 残差边
  Identity Block模块残差边不进行任何操作，将第三次卷积后的结果与输入相加，然后经过激活函数即得到输出结果。

x = layers.add([x, input_tensor])
x = Activation('relu')(x)
return x

---

ResNet50模型就是将很多个Conv Block和Identity Block进行堆叠，最后得到的输出矩阵大小为[7, 7, 2048]，对该矩阵使用kernel_size为7的pooling即可将矩阵变为1维，接着节Dense层进行分类预测。