深度残差网络（ResNet）详解与实现（tensorflow2.x）

ResNet原理

深层网络在学习任务中取得了超越人眼的准确率，但是，经过实验表明，模型的性能和模型的深度并非成正比，是由于模型的表达能力过强，反而在测试数据集中性能下降。ResNet的核心是，为了防止梯度弥散或爆炸，让信息流经快捷连接到达浅层。
更正式的讲，输入$x$通过卷积层,得到特征变换后的输出$F(x)$,与输入$x$进行对应元素的相加运算,得到最终输出$H(x)$:
$$H(x)=x+F(x)$$
VGG模块和残差模块对比如下：

为了能够满足输入$x$与卷积层的输出$F(x)$能够相加运算,需要输入$x$的 shape 与$F(x)$的shape 完全一致。当出现 shape 不一致时，一般通过Conv2D进行变换，该Conv2D的核为1×1，步幅为2。

ResNet实现

使用tensorflow2.3实现ResNet

模型创建

import numpy as np
import tensorflow as tf
from tensorflow import keras
from matplotlib import pyplot as plt
import os
import math
"""
用于控制模型层数
"""
#残差块数
n = 3
depth = n * 9 + 1
def resnet_layer(inputs,
        num_filters=16,
        kernel_size=3,
        strides=1,
        activation='relu',
        batch_normalization=True,
        conv_first=True):
    """2D Convolution-Batch Normalization-Activation stack builder
    Arguments:
        inputs (tensor): 输入
        num_filters (int): 卷积核个数
        kernel_size (int): 卷积核大小
        activation (string): 激活层
        batch_normalization (bool): 是否使用批归一化
        conv_first (bool): conv-bn-active(True) or bn-active-conv (False)层堆叠次序
    
    Returns:
        x (tensor): 输出
    """
    conv = keras.layers.Conv2D(num_filters,
            kernel_size=kernel_size,
            strides=strides,
            padding='same',
            kernel_initializer='he_normal',
            kernel_regularizer=keras.regularizers.l2(1e-4))
    x = inputs
    if conv_first:
        x = conv(x)
        if batch_normalization:
            x = keras.layers.BatchNormalization()(x)
        if activation is not None:
            x = keras.layers.Activation(activation)(x)
    else:
        if batch_normalization:
            x = keras.layers.BatchNormalization()(x)
        if activation is not None:
            x = keras.layers.Activation(activation)(x)
        x = conv(x)
    return x


def resnet(input_shape,depth,num_classes=10):
    """ResNet

    Arguments:
        input_shape (tensor): 输入尺寸
        depth (int): 网络层数
        num_classes (int): 预测类别数
    
    Return:
        model (Model): 模型
    """
    if (depth - 2) % 6 != 0:
        raise ValueError('depth should be 6n+2')
    #超参数
    num_filters = 16
    num_res_blocks = int((depth - 2) / 6)

    inputs = keras.layers.Input(shape=input_shape)
    x = resnet_layer(inputs=inputs)
    for stack in range(3):
        for res_block in range(num_res_blocks):
            strides = 1
            if stack > 0 and res_block == 0:
                strides = 2
            y = resnet_layer(inputs=x,num_filters=num_filters,
                    strides=strides)
            y = resnet_layer(inputs=y,num_filters=num_filters,
                    activation=None)
            if stack > 0 and res_block == 0:
                x = resnet_layer(inputs=x,
                        num_filters=num_filters,
                        kernel_size=1,
                        strides=strides,
                        activation=None,
                        batch_normalization=False)
            x = keras.layers.add([x,y])
            x = keras.layers.Activation('relu')(x)
        num_filters *= 2
    x = keras.layers.AveragePooling2D(pool_size=8)(x)
    x = keras.layers.Flatten()(x)
    outputs = keras.layers.Dense(num_classes,activation='softmax',
            kernel_initializer='he_normal')(x)
    model = keras.Model(inputs=inputs,outputs=outputs)
    return model

model = resnet_v1(input_shape=input_shape,depth=depth)

数据加载

#加载数据
(x_train,y_train),(x_test,y_test) = keras.datasets.cifar10.load_data()

#计算类别数
num_labels = len(np.unique(y_train))

#转化为one-hot编码
y_train = keras.utils.to_categorical(y_train)
y_test = keras.utils.to_categorical(y_test)

#预处理
input_shape = x_train.shape[1:]
x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.

模型编译

#超参数
batch_size = 64
epochs = 200
#编译模型
model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['acc'])
model.summary()

模型训练

model.fit(x_train,y_train,
        batch_size=batch_size,
        epochs=epochs,
        validation_data=(x_test,y_test),
        shuffle=True)

测试模型

scores = model.evaluate(x_test,y_test,batch_size=batch_size,verbose=0)
print('Test loss: ',scores[0])
print('Test accuracy: ',scores[1])

训练过程

Epoch 104/200
782/782 [==============================] - ETA: 0s - loss: 0.2250 - acc: 0.9751 
Epoch 00104: val_acc did not improve from 0.91140
782/782 [==============================] - 15s 19ms/step - loss: 0.2250 - acc: 0.9751 - val_loss: 0.4750 - val_acc: 0.9090
learning rate:  0.0001
Epoch 105/200
781/782 [============================>.] - ETA: 0s - loss: 0.2206 - acc: 0.9754 
Epoch 00105: val_acc did not improve from 0.91140
782/782 [==============================] - 16s 20ms/step - loss: 0.2206 - acc: 0.9754 - val_loss: 0.4687 - val_acc: 0.9078
learning rate:  0.0001
Epoch 106/200
782/782 [==============================] - ETA: 0s - loss: 0.2160 - acc: 0.9769 
Epoch 00106: val_acc did not improve from 0.91140
782/782 [==============================] - 15s 20ms/step - loss: 0.2160 - acc: 0.9769 - val_loss: 0.4886 - val_acc: 0.9053