keras迁移学习猫狗大战-Vgg16

VGG16模型

转载vgg16-Bubbliiiing
VGG是由Simonyan 和Zisserman在文献《Very Deep Convolutional Networks for Large Scale Image Recognition》中提出卷积神经网络模型，其名称来源于作者所在的牛津大学视觉几何组(Visual Geometry Group)的缩写。
该模型参加2014年的 ImageNet图像分类与定位挑战赛，取得了优异成绩：在分类任务上排名第二，在定位任务上排名第一。
可能大家会想，这样一个这么强的模型肯定很复杂吧？

它的结构如下图所示：

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         (None, 224, 224, 3)       0
_________________________________________________________________
block1_conv1 (Conv2D)        (None, 224, 224, 64)      1792
_________________________________________________________________
block1_conv2 (Conv2D)        (None, 224, 224, 64)      36928
_________________________________________________________________
block1_pool (MaxPooling2D)   (None, 112, 112, 64)      0
_________________________________________________________________
block2_conv1 (Conv2D)        (None, 112, 112, 128)     73856
_________________________________________________________________
block2_conv2 (Conv2D)        (None, 112, 112, 128)     147584
_________________________________________________________________
block2_pool (MaxPooling2D)   (None, 56, 56, 128)       0
_________________________________________________________________
block3_conv1 (Conv2D)        (None, 56, 56, 256)       295168
_________________________________________________________________
block3_conv2 (Conv2D)        (None, 56, 56, 256)       590080
_________________________________________________________________
block3_conv3 (Conv2D)        (None, 56, 56, 256)       590080
_________________________________________________________________
block3_pool (MaxPooling2D)   (None, 28, 28, 256)       0
_________________________________________________________________
block4_conv1 (Conv2D)        (None, 28, 28, 512)       1180160
_________________________________________________________________
block4_conv2 (Conv2D)        (None, 28, 28, 512)       2359808
_________________________________________________________________
block4_conv3 (Conv2D)        (None, 28, 28, 512)       2359808
_________________________________________________________________
block4_pool (MaxPooling2D)   (None, 14, 14, 512)       0
_________________________________________________________________
block5_conv1 (Conv2D)        (None, 14, 14, 512)       2359808
_________________________________________________________________
block5_conv2 (Conv2D)        (None, 14, 14, 512)       2359808
_________________________________________________________________
block5_conv3 (Conv2D)        (None, 14, 14, 512)       2359808
_________________________________________________________________
block5_pool (MaxPooling2D)   (None, 7, 7, 512)         0
_________________________________________________________________
flatten (Flatten)            (None, 25088)             0
_________________________________________________________________
fullc1 (Dense)               (None, 4096)              102764544
_________________________________________________________________
fullc2 (Dense)               (None, 4096)              16781312
fullc3 (Dense)               (None, 20)                81940
=================================================================
Total params: 134,342,484
Trainable params: 134,342,484
Non-trainable params: 0
_________________________________________________________________
None

因为只对猫和狗进行分类，所以最后的全连接层可以缩小一点，本例子中将其缩小到256。
前面的卷积层可以不训练，因为卷积层的作用主要是提取特征，已经训练好的VGG提取的特征都是比较有用的，所以可以不进行调整，只对最后三个全连接层进行训练。

训练前的准备

1、数据集处理

在数据集处理之前，首先要下载猫狗数据集，地址如下。
链接：https://pan.baidu.com/s/1HBewIgKsFD8hh3ICOnnTwA
提取码：ktab
源代码
链接: https://pan.baidu.com/s/1r79Ey_2ZPZc73GND-kUxxg
提取码: w5bg
下载完后数据集如下，我们需要将其记录在文本中可以让后续可以进行处理

数据存储在train中，我们运行generat_traintxt.py生成train.txt
generat_traintxt.py
就是列举文件夹中的文件，然后根据文件名在末尾打上标签

import os

with open('./train.txt','w') as f:
    after_generate = os.listdir("./train")
    for image in after_generate:
        if image.split(".")[0]=='cat':
            f.write(image + ";" + "0" + "\n")
        else:
            f.write(image + ";" + "1" + "\n")

2创建vgg16 Keras模型

VGG的模型的结构已经在上面介绍过了，利用Keras就可以构建。在这里我给他添加了一个7x7的卷积层，同时把全连接层改为256。

import tensorflow as tf
from tensorflow import keras
from keras import Model,Sequential
from keras.layers import Flatten, Dense, Conv2D, GlobalAveragePooling2D
from keras.layers import Input, MaxPooling2D, GlobalMaxPooling2D


def VGG16(num_classes):

    image_input = Input(shape = (224,224,3))
    # 第一个卷积部分
    x = Conv2D(64,(3,3),activation = 'relu',padding = 'same',name = 'block1_conv1')(image_input)
    x = Conv2D(64,(3,3),activation = 'relu',padding = 'same', name = 'block1_conv2')(x)
    x = MaxPooling2D((2,2), strides = (2,2), name = 'block1_pool')(x)

    # 第二个卷积部分
    x = Conv2D(128,(3,3),activation = 'relu',padding = 'same',name = 'block2_conv1')(x)
    x = Conv2D(128,(3,3),activation = 'relu',padding = 'same',name = 'block2_conv2')(x)
    x = MaxPooling2D((2,2),strides = (2,2),name = 'block2_pool')(x)

    # 第三个卷积部分
    x = Conv2D(256,(3,3),activation = 'relu',padding = 'same',name = 'block3_conv1')(x)
    x = Conv2D(256,(3,3),activation = 'relu',padding = 'same',name = 'block3_conv2')(x)
    x = Conv2D(256,(3,3),activation = 'relu',padding = 'same',name = 'block3_conv3')(x)
    x = MaxPooling2D((2,2),strides = (2,2),name = 'block3_pool')(x)

    # 第四个卷积部分
    x = Conv2D(512,(3,3),activation = 'relu',padding = 'same', name = 'block4_conv1')(x)
    x = Conv2D(512,(3,3),activation = 'relu',padding = 'same', name = 'block4_conv2')(x)
    x = Conv2D(512,(3,3),activation = 'relu',padding = 'same', name = 'block4_conv3')(x)
    x = MaxPooling2D((2,2),strides = (2,2),name = 'block4_pool')(x)

    # 第五个卷积部分
    x = Conv2D(512,(3,3),activation = 'relu',padding = 'same', name = 'block5_conv1')(x)
    x = Conv2D(512,(3,3),activation = 'relu',padding = 'same', name = 'block5_conv2')(x)
    x = Conv2D(512,(3,3),activation = 'relu',padding = 'same', name = 'block5_conv3')(x)
    x = MaxPooling2D((2,2),strides = (2,2),name = 'block5_pool')(x)

    # 分类部分
    x = Conv2D(256,(7,7),activation = 'relu',padding = 'valid', name = 'block6_conv4')(x)
    x = Flatten(name = 'flatten')(x)
    x = Dense(256,activation = 'relu',name = 'fullc1')(x)
    x = Dense(256,activation = 'relu',name = 'fullc2')(x)
    x = Dense(num_classes,activation = 'softmax',name = 'fullc3')(x)
    model = Model(image_input,x,name = 'vgg16')

    return model

下载权重

这个网址拉到最下面就可以了。
https://github.com/fchollet/deep-learning-models/releases
下载这个文件：vgg16_weights_tf_dim_ordering_tf_kernels.h5。

开始训练

训练的主函数主要包括如下部分：
1、读取训练用txt，并打乱，利用该txt进行训练集和测试集的划分。
2、建立VGG16模型，载入权重。这里要注意skip_mismatch=True。
3、利用model.layers[i].trainable = False将VGG16前面的卷积层设置成不可训练。仅训练最后2层。
3、设定模型保存的方式、学习率下降的方式、是否需要早停。
4、利用model.fit_generator训练模型。
具体代码如下：

from keras.callbacks import TensorBoard, ModelCheckpoint, ReduceLROnPlateau, EarlyStopping
from keras.utils import np_utils
from keras.optimizers import Adam
from vgg import VGG16
import numpy as np
import utils
import cv2
from keras import backend as K

K.set_image_data_format('channels_last')


def generate_arrays_from_file(lines, batch_size):
    # 获取总长度
    n = len(lines)
    i = 0
    while 1:
        X_train = []
        Y_train = []
        # 获取一个batch_size大小的数据
        for b in range(batch_size):
            if i == 0:
                np.random.shuffle(lines)
            name = lines[i].split(';')[0]
            # 从文件中读取图像
            img = cv2.imread("./vgg/data/train" + '/' + name)
            img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
            img = img / 255
            X_train.append(img)
            Y_train.append(lines[i].split(';')[1])
            # 读完一个周期后重新开始
            i = (i + 1) % n
        # 处理图像
        X_train = utils.resize_image(X_train, (224, 224))
        X_train = X_train.reshape(-1, 224, 224, 3)
        Y_train = np_utils.to_categorical(np.array(Y_train), num_classes=2)
        yield (X_train, Y_train)


if __name__ == "__main__":
    # 模型保存的位置
    log_dir = "./vgg/logs/"

    # 打开数据集的txt
    with open("./vgg/data/train.txt", "r") as f:
        lines = f.readlines()

    # 打乱行，这个txt主要用于帮助读取数据来训练
    # 打乱的数据更有利于训练
    np.random.seed(10101)
    np.random.shuffle(lines)
    np.random.seed(None)

    # 90%用于训练，10%用于估计。
    num_val = int(len(lines) * 0.1)
    num_train = len(lines) - num_val

    # 建立AlexNet模型
    model = VGG16(2)

    # 注意要开启skip_mismatch和by_name
    model.load_weights("./vgg/vgg16_weights_tf_dim_ordering_tf_kernels.h5", by_name=True, skip_mismatch=True)

    # 指定训练层
    for i in range(0, len(model.layers) - 2):
        model.layers[i].trainable = False

    # 保存的方式，3世代保存一次
    checkpoint_period1 = ModelCheckpoint(
        log_dir + 'ep{epoch:03d}-loss{loss:.3f}-val_loss{val_loss:.3f}.h5',
        monitor='acc',
        save_weights_only=False,
        save_best_only=True,
        period=3
    )
    # 学习率下降的方式，acc三次不下降就下降学习率继续训练
    reduce_lr = ReduceLROnPlateau(
        monitor='acc',
        factor=0.5,
        patience=3,
        verbose=1
    )
    # 是否需要早停，当val_loss一直不下降的时候意味着模型基本训练完毕，可以停止
    early_stopping = EarlyStopping(
        monitor='val_loss',
        min_delta=0,
        patience=10,
        verbose=1
    )

    # 交叉熵
    model.compile(loss='categorical_crossentropy',
                  optimizer=Adam(lr=1e-3),
                  metrics=['accuracy'])

    # 一次的训练集大小
    batch_size = 16

    print('Train on {} samples, val on {} samples, with batch size {}.'.format(num_train, num_val, batch_size))

    # 开始训练
    model.fit_generator(generate_arrays_from_file(lines[:num_train], batch_size),
                        steps_per_epoch=max(1, num_train // batch_size),
                        validation_data=generate_arrays_from_file(lines[num_train:], batch_size),
                        validation_steps=max(1, num_val // batch_size),
                        epochs=50,
                        initial_epoch=0,
                        callbacks=[checkpoint_period1, reduce_lr])

    model.save_weights(log_dir + 'last1.h5')

在训练和预测中用到了util模块的代码调用，具体为裁减图形 重新设定输入图片大小为（n，224，224，3），

import matplotlib.image as mpimg
import numpy as np
import cv2
import tensorflow as tf
from tensorflow.python.ops import array_ops


def load_image(path):
    # 读取图片，rgb
    img = mpimg.imread(path)
    # 将图片修剪成中心的正方形
    short_edge = min(img.shape[:2])
    yy = int((img.shape[0] - short_edge) / 2)
    xx = int((img.shape[1] - short_edge) / 2)
    crop_img = img[yy: yy + short_edge, xx: xx + short_edge]
    return crop_img


def resize_image(image, size):
    with tf.name_scope('resize_image'):
        images = []
        for i in image:
            i = cv2.resize(i, size)
            images.append(i)
        images = np.array(images)
        return images

结果预测

具体说明图像显示

mport numpy as np
import utils
import cv2
from keras import backend as K
from vgg import VGG16
import matplotlib.pyplot as plt

K.set_image_data_format('channels_last')
class_classification = ["cat", "dog"]

if __name__ == "__main__":
    model = VGG16(2)
    model.load_weights("./logs/last1.h5")
    # img = cv2.imread("./data/train/cat.1.jpg")
    # img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
    # img = img/255
    # img = np.expand_dims(img,axis = 0)
    # img = utils.resize_image(img,(224,224))
    # utils.print_answer(np.argmax(model.predict(img)))
    # print(utils.print_answer(np.argmax(model.predict(img))))
    # print(class_classification[np.argmax(model.predict(img))])
    # show
    with open("./data/train.txt", "r") as f:
        lines = f.readlines()
        np.random.shuffle(lines)
    plt.figure(figsize=(10, 10))
    for i in range(9):
        plt.subplot(3, 3, i+1)
        num = int(np.random.random() * len(lines))
        name = lines[num].split(';')[0]
        img = cv2.imread("./data/train" + '/' + name)
        img_input = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        img_input = img_input / 255
        print("image=",np.shape(img_input))
        img_input = np.expand_dims(img_input, axis=0)
        print("image2=",np.shape(img_input))
        img_input = utils.resize_image(img_input, (224, 224))
        print("image3=" , np.shape(img_input))
        predit = class_classification[np.argmax(model.predict(img_input))]
        plt.title("label=%s predit=%s"%(name.split(".")[0], predit))
        plt.imshow(cv2.resize(img,(224,224)))
        plt.axis('off')
    plt.show()