一、简介
- 迁移学习：一种非常强大的深度学习技术，它的动机很简单——“站在巨人肩膀上”。假设你想学习一门西班牙语，如果从已经掌握的英语学习便更加简单高效。
- VGG16：2014年牛津大学提出的模型，简洁实用，擅长图像分类和目标检测。
- 猫狗大战数据集：Kaggle五年前提供的数据集，含25000张狗和猫的图像（共50000张），另有用于测试的12500张。https://www.kaggle.com/c/dogs-vs-cats/data
二、训练代码
备注：
-
建议分步用IPython运行。
-
训练数据集用的是自定义目录，请自行更改。
我的目录结构如下：
-
训练样本数和验证样本数可根据电脑配置和时间调整。下载官网数据集后把train.zip解压后根据数目随意分配到对应文件夹下即可。
from keras import optimizers
from keras import applications
from keras.models import Sequential, Model
from keras.callbacks import ModelCheckpoint
from keras.layers import Dropout, Flatten, Dense
from keras.preprocessing.image import ImageDataGenerator
# 数据集
img_height, img_width = 256, 256 # 图片高宽
batch_size = 2 # 批量大小
epochs = 50 # 迭代次数
train_data_dir = 'data/dogs_and_cats/train' # 训练集目录
validation_data_dir = 'data/dogs_and_cats/validation' # 测试集目录
OUT_CATEGORIES = 1 # 分类数
nb_train_samples = 2000 # 训练样本数
nb_validation_samples = 200 # 验证样本数
# 定义模型
base_model = applications.VGG16(weights="imagenet", include_top=False,
input_shape=(img_width, img_height, 3)) # 预训练的VGG16网络，替换掉顶部网络
print(base_model.summary())
for layer in base_model.layers[:15]: layer.trainable = False # 冻结预训练网络前15层
top_model = Sequential() # 自定义顶层网络
top_model.add(Flatten(input_shape=base_model.output_shape[1:])) # 将预训练网络展平
top_model.add(Dense(256, activation='relu')) # 全连接层，输入像素256
top_model.add(Dropout(0.5)) # Dropout概率0.5
top_model.add(Dense(OUT_CATEGORIES, activation='sigmoid')) # 输出层，二分类
# top_model.load_weights("") # 单独训练的自定义网络
model = Model(inputs=base_model.input, outputs=top_model(base_model.output)) # 新网络=预训练网络+自定义网络
model.compile(loss='binary_crossentropy', optimizer=optimizers.SGD(lr=0.0001, momentum=0.9),
metrics=['accuracy']) # 损失函数为二进制交叉熵，优化器为SGD
train_datagen = ImageDataGenerator(rescale=1. / 255, horizontal_flip=True) # 训练数据预处理器，随机水平翻转
test_datagen = ImageDataGenerator(rescale=1. / 255) # 测试数据预处理器
train_generator = train_datagen.flow_from_directory(train_data_dir, target_size=(img_height, img_width),
batch_size=batch_size, class_mode='binary') # 训练数据生成器
validation_generator = test_datagen.flow_from_directory(validation_data_dir, target_size=(img_height, img_width),
batch_size=batch_size, class_mode='binary',
shuffle=False) # 验证数据生成器
checkpointer = ModelCheckpoint(filepath='dogcatmodel.h5', verbose=1, save_best_only=True) # 保存最优模型
# 训练&评估
model.fit_generator(train_generator, steps_per_epoch=nb_train_samples // batch_size, epochs=epochs,
validation_data=validation_generator, validation_steps=nb_validation_samples // batch_size,
verbose=2, workers=12, callbacks=[checkpointer]) # 每轮一行输出结果，最大进程12
三、训练结果
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) (None, 256, 256, 3) 0
_________________________________________________________________
block1_conv1 (Conv2D) (None, 256, 256, 64) 1792
_________________________________________________________________
block1_conv2 (Conv2D) (None, 256, 256, 64) 36928
_________________________________________________________________
block1_pool (MaxPooling2D) (None, 128, 128, 64) 0
_________________________________________________________________
block2_conv1 (Conv2D) (None, 128, 128, 128) 73856
_________________________________________________________________
block2_conv2 (Conv2D) (None, 128, 128, 128) 147584
_________________________________________________________________
block2_pool (MaxPooling2D) (None, 64, 64, 128) 0
_________________________________________________________________
block3_conv1 (Conv2D) (None, 64, 64, 256) 295168
_________________________________________________________________
block3_conv2 (Conv2D) (None, 64, 64, 256) 590080
_________________________________________________________________
block3_conv3 (Conv2D) (None, 64, 64, 256) 590080
_________________________________________________________________
block3_pool (MaxPooling2D) (None, 32, 32, 256) 0
_________________________________________________________________
block4_conv1 (Conv2D) (None, 32, 32, 512) 1180160
_________________________________________________________________
block4_conv2 (Conv2D) (None, 32, 32, 512) 2359808
_________________________________________________________________
block4_conv3 (Conv2D) (None, 32, 32, 512) 2359808
_________________________________________________________________
block4_pool (MaxPooling2D) (None, 16, 16, 512) 0
_________________________________________________________________
block5_conv1 (Conv2D) (None, 16, 16, 512) 2359808
_________________________________________________________________
block5_conv2 (Conv2D) (None, 16, 16, 512) 2359808
_________________________________________________________________
block5_conv3 (Conv2D) (None, 16, 16, 512) 2359808
_________________________________________________________________
block5_pool (MaxPooling2D) (None, 8, 8, 512) 0
=================================================================
Total params: 14,714,688
Trainable params: 14,714,688
Non-trainable params: 0
_________________________________________________________________
None
Found 24800 images belonging to 2 classes.
Found 200 images belonging to 2 classes.
省略一堆
Epoch 49/50
- 47s - loss: 0.0412 - acc: 0.9840 - val_loss: 0.0477 - val_acc: 0.9850
Epoch 00049: val_loss improved from 0.05592 to 0.04766, saving model to dogcatmodel.h5
得到模型
四、预测代码
from keras.models import load_model
from keras.preprocessing import image
# 图片预处理
path = 'data/dogs_and_cats/test1/1.jpg' # 图片路径
img_height, img_width = 256, 256 # 图片宽高
x = image.load_img(path=path, target_size=(img_height, img_width)) # 加载图片
x = image.img_to_array(x) # 图片转ndarray
x = x[None] # 升维
# 预测
model = load_model('dogcatmodel.h5') # 加载模型
y = model.predict(x)
y = 'cat' if y == 0 else 'dog' # 0猫1狗
print(y)
五、预测结果
备注：该图为print输出后的人工标注，非机器标注。
那么问题来了，我家俩肥猫是猫吗？
