本文为为🔗365天深度学习训练营内部文章
原作者：K同学啊
一 导入数据
python
import matplotlib.pyplot as plt
import tensorflow as tf
# 支持中文
plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号
import os,PIL,pathlib
#隐藏警告
import warnings
warnings.filterwarnings('ignore')
data_dir = "./data"
data_dir = pathlib.Path(data_dir)
image_count = len(list(data_dir.glob('*/*')))
print("图片总数为：",image_count)图片总数为： 3400二、数据预处理
1. 加载数据
使用
image_dataset_from_directory方法将磁盘中的数据加载到tf.data.Dataset中
python
batch_size = 8
img_height = 224
img_width = 224
python
train_ds = tf.keras.preprocessing.image_dataset_from_directory(
data_dir,
validation_split=0.2,
subset="training",
seed=12,
image_size=(img_height, img_width),
batch_size=batch_size)Found 3400 files belonging to 2 classes.
Using 2720 files for training.
python
val_ds = tf.keras.preprocessing.image_dataset_from_directory(
data_dir,
validation_split=0.2,
subset="validation",
seed=12,
image_size=(img_height, img_width),
batch_size=batch_size)Found 3400 files belonging to 2 classes.
Using 680 files for validation.
python
class_names = train_ds.class_names
print(class_names)['cat', 'dog']2. 再次检查数据
python
for image_batch, labels_batch in train_ds:
print(image_batch.shape)
print(labels_batch.shape)
break(8, 224, 224, 3)
(8,)3. 配置数据集
- shuffle() ： 打乱数据
- prefetch() ：预取数据，加速运行
- cache() ：将数据集缓存到内存当中，加速运行
python
AUTOTUNE = tf.data.AUTOTUNE
def preprocess_image(image,label):
return (image/255.0,label)
# 归一化处理
train_ds = train_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)
val_ds = val_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)
train_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)在这里数据处理的过程中，比前几次稍微不同的是多加了一个归一化的处理
4.可视化数据
python
plt.figure(figsize=(15, 10)) # 图形的宽为15高为10
for images, labels in train_ds.take(1):
for i in range(8):
ax = plt.subplot(5, 8, i + 1)
plt.imshow(images[i])
plt.title(class_names[labels[i]])
plt.axis("off")
三、构建VG-16网络
VGG优缺点分析：
- VGG优点
VGG的结构非常简洁，整个网络都使用了同样大小的卷积核尺寸（3x3）和最大池化尺寸（2x2）。
- VGG缺点
1)训练时间过长，调参难度大。2)需要的存储容量大，不利于部署。例如存储VGG-16权重值文件的大小为500多MB，不利于安装到嵌入式系统中。
结构说明：
- 13个卷积层（Convolutional Layer），分别用
blockX_convX表示
- 3个全连接层（Fully connected Layer），分别用
fcX与predictions表示
- 5个池化层（Pool layer），分别用
blockX_pool表示VGG-16****包含了16个隐藏层（13个卷积层和3个全连接层），故称为 VGG-16
构建方法1：调用官网封装好的模型函数
python
model = tf.keras.applications.VGG16(weights='imagenet')
model.summary()Model: "vgg16"
_________________________________________________________________
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
fc1 (Dense) (None, 4096) 102764544
fc2 (Dense) (None, 4096) 16781312
predictions (Dense) (None, 1000) 4097000
=================================================================
Total params: 138,357,544
Trainable params: 138,357,544
Non-trainable params: 0
_________________________________________________________________构建方法二：自己手动搭建模型
python
from tensorflow.keras import layers, models, Input
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Dense, Flatten, Dropout
def VGG16(nb_classes, input_shape):
input_tensor = Input(shape=input_shape)
# 1st block
x = Conv2D(64, (3,3), activation='relu', padding='same',name='block1_conv1')(input_tensor)
x = Conv2D(64, (3,3), activation='relu', padding='same',name='block1_conv2')(x)
x = MaxPooling2D((2,2), strides=(2,2), name = 'block1_pool')(x)
# 2nd block
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)
# 3rd block
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)
# 4th block
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)
# 5th block
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)
# full connection
x = Flatten()(x)
x = Dense(4096, activation='relu', name='fc1')(x)
x = Dense(4096, activation='relu', name='fc2')(x)
output_tensor = Dense(nb_classes, activation='softmax', name='predictions')(x)
model = Model(input_tensor, output_tensor)
return model
model=VGG16(1000, (img_width, img_height, 3))
model.summary()Model: "model"
_________________________________________________________________
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
fc1 (Dense) (None, 4096) 102764544
fc2 (Dense) (None, 4096) 16781312
predictions (Dense) (None, 1000) 4097000
=================================================================
Total params: 138,357,544
Trainable params: 138,357,544
Non-trainable params: 0
_________________________________________________________________四、编译
在准备对模型进行训练之前，还需要再对其进行一些设置。以下内容是在模型的编译步骤中添加的：
- 损失函数（loss）：用于衡量模型在训练期间的准确率。
- 优化器（optimizer）：决定模型如何根据其看到的数据和自身的损失函数进行更新。
- 评价函数（metrics）：用于监控训练和测试步骤。以下示例使用了准确率，即被正确分类的图像的比率。
python
# 设置初始学习率
initial_learning_rate = 1e-4
lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(
initial_learning_rate,
decay_steps=30, # 敲黑板！！！这里是指 steps，不是指epochs
decay_rate=0.92, # lr经过一次衰减就会变成 decay_rate*lr
staircase=True)
# 设置优化器
opt = tf.keras.optimizers.Adam(learning_rate=initial_learning_rate)
model.compile(optimizer=opt,
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])早停法：
python
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping
epochs = 10
# 保存最佳模型参数
checkpointer = ModelCheckpoint('best_model.h5',
monitor='val_accuracy',
verbose=1,
save_best_only=True,
save_weights_only=True)
# 设置早停
earlystopper = EarlyStopping(monitor='val_accuracy',
min_delta=0.001,
patience=20,
verbose=1)五、训练模型
python
history = model.fit(train_ds,
validation_data=val_ds,
epochs=epochs,
callbacks=[checkpointer, earlystopper])Epoch 1/10
340/340 [==============================] - ETA: 0s - loss: 0.2396 - accuracy: 0.8930
Epoch 1: val_accuracy improved from -inf to 0.99412, saving model to best_model.h5
340/340 [==============================] - 1376s 4s/step - loss: 0.2396 - accuracy: 0.8930 - val_loss: 0.0210 - val_accuracy: 0.9941
Epoch 2/10
340/340 [==============================] - ETA: 0s - loss: 0.0276 - accuracy: 0.9908
Epoch 2: val_accuracy did not improve from 0.99412
340/340 [==============================] - 1345s 4s/step - loss: 0.0276 - accuracy: 0.9908 - val_loss: 0.0465 - val_accuracy: 0.9853
Epoch 3/10
340/340 [==============================] - ETA: 0s - loss: 0.1150 - accuracy: 0.9717
Epoch 3: val_accuracy did not improve from 0.99412
340/340 [==============================] - 1316s 4s/step - loss: 0.1150 - accuracy: 0.9717 - val_loss: 0.0704 - val_accuracy: 0.9750
Epoch 4/10
340/340 [==============================] - ETA: 0s - loss: 0.0192 - accuracy: 0.9949
Epoch 4: val_accuracy improved from 0.99412 to 0.99853, saving model to best_model.h5
340/340 [==============================] - 1336s 4s/step - loss: 0.0192 - accuracy: 0.9949 - val_loss: 0.0083 - val_accuracy: 0.9985
Epoch 5/10
340/340 [==============================] - ETA: 0s - loss: 0.0248 - accuracy: 0.9930
Epoch 5: val_accuracy did not improve from 0.99853
340/340 [==============================] - 1321s 4s/step - loss: 0.0248 - accuracy: 0.9930 - val_loss: 0.0036 - val_accuracy: 0.9985
Epoch 6/10
340/340 [==============================] - ETA: 0s - loss: 0.0240 - accuracy: 0.9937
Epoch 6: val_accuracy did not improve from 0.99853
340/340 [==============================] - 1323s 4s/step - loss: 0.0240 - accuracy: 0.9937 - val_loss: 0.0074 - val_accuracy: 0.9956
Epoch 7/10
340/340 [==============================] - ETA: 0s - loss: 0.0039 - accuracy: 0.9982
Epoch 7: val_accuracy did not improve from 0.99853
340/340 [==============================] - 1324s 4s/step - loss: 0.0039 - accuracy: 0.9982 - val_loss: 0.0069 - val_accuracy: 0.9971
Epoch 8/10
340/340 [==============================] - ETA: 0s - loss: 8.3202e-04 - accuracy: 1.0000
Epoch 8: val_accuracy did not improve from 0.99853
340/340 [==============================] - 1318s 4s/step - loss: 8.3202e-04 - accuracy: 1.0000 - val_loss: 0.0205 - val_accuracy: 0.9956
Epoch 9/10
340/340 [==============================] - ETA: 0s - loss: 0.0759 - accuracy: 0.9801
Epoch 9: val_accuracy did not improve from 0.99853
340/340 [==============================] - 1326s 4s/step - loss: 0.0759 - accuracy: 0.9801 - val_loss: 0.0372 - val_accuracy: 0.9882
Epoch 10/10
340/340 [==============================] - ETA: 0s - loss: 0.0242 - accuracy: 0.9934
Epoch 10: val_accuracy did not improve from 0.99853
340/340 [==============================] - 1328s 4s/step - loss: 0.0242 - accuracy: 0.9934 - val_loss: 0.0072 - val_accuracy: 0.9985六 模型评估
python
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs_range = range(epochs)
plt.figure(figsize=(12, 4))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, acc, label='Training Accuracy')
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
预测
python
import numpy as np
# 采用加载的模型（new_model）来看预测结果
plt.figure(figsize=(18, 3)) # 图形的宽为18高为5
plt.suptitle("预测结果展示")
for images, labels in val_ds.take(1):
for i in range(8):
ax = plt.subplot(1,8, i + 1)
# 显示图片
plt.imshow(images[i].numpy())
# 需要给图片增加一个维度
img_array = tf.expand_dims(images[i], 0)
# 使用模型预测图片中的人物
predictions = model.predict(img_array)
plt.title(class_names[np.argmax(predictions)])
plt.axis("off")1/1 [==============================] - 1s 609ms/step
1/1 [==============================] - 0s 123ms/step
1/1 [==============================] - 0s 140ms/step
1/1 [==============================] - 0s 134ms/step
1/1 [==============================] - 0s 129ms/step
1/1 [==============================] - 0s 126ms/step
1/1 [==============================] - 0s 124ms/step
1/1 [==============================] - 0s 123ms/step
在训练模型的时候，除了用上述的代码之外，还可以用另一种方式。
改用model.train_on_batch方法。两种方法的比较：
model.fit()：用起来十分简单，对新手非常友好
model.train_on_batch()：封装程度更低，可以玩更多花样。此外我也引入了进度条的显示方式，更加方便我们及时查看模型训练过程中的情况，可以及时打印各项指标
python
model.compile(optimizer="adam",
loss ='sparse_categorical_crossentropy',
metrics =['accuracy'])
python
from tqdm import tqdm
import tensorflow.keras.backend as K
epochs = 10
lr = 1e-4
# 记录训练数据，方便后面的分析
history_train_loss = []
history_train_accuracy = []
history_val_loss = []
history_val_accuracy = []
for epoch in range(epochs):
train_total = len(train_ds)
val_total = len(val_ds)
"""
total：预期的迭代数目
ncols：控制进度条宽度
mininterval：进度更新最小间隔，以秒为单位（默认值：0.1）
"""
with tqdm(total=train_total, desc=f'Epoch {epoch + 1}/{epochs}',mininterval=1,ncols=100) as pbar:
lr = lr*0.92
K.set_value(model.optimizer.lr, lr)
for image,label in train_ds:
"""
训练模型，简单理解train_on_batch就是：它是比model.fit()更高级的一个用法
想详细了解 train_on_batch 的同学，
可以看看我的这篇文章：https://www.yuque.com/mingtian-fkmxf/hv4lcq/ztt4gy
"""
history = model.train_on_batch(image,label)
train_loss = history[0]
train_accuracy = history[1]
pbar.set_postfix({"loss": "%.4f"%train_loss,
"accuracy":"%.4f"%train_accuracy,
"lr": K.get_value(model.optimizer.lr)})
pbar.update(1)
history_train_loss.append(train_loss)
history_train_accuracy.append(train_accuracy)
print('开始验证！')
with tqdm(total=val_total, desc=f'Epoch {epoch + 1}/{epochs}',mininterval=0.3,ncols=100) as pbar:
for image,label in val_ds:
history = model.test_on_batch(image,label)
val_loss = history[0]
val_accuracy = history[1]
pbar.set_postfix({"loss": "%.4f"%val_loss,
"accuracy":"%.4f"%val_accuracy})
pbar.update(1)
history_val_loss.append(val_loss)
history_val_accuracy.append(val_accuracy)
print('结束验证！')
print("验证loss为：%.4f"%val_loss)
print("验证准确率为：%.4f"%val_accuracy)
对比之前的model.fit()方法，这次还引用了更详细的进度条。后续的操作和上述方法一样