>- **🍨 本文为[🔗365天深度学习训练营](https://mp.weixin.qq.com/s/k-vYaC8l7uxX51WoypLkTw) 中的学习记录博客**
>- **🍦 参考文章:365天深度学习训练营-第6周:好莱坞明星识别(训练营内部成员可读)**
>- **🍖 原作者:[K同学啊|接辅导、项目定制](https://mtyjkh.blog.csdn.net/)**
目录
一 前期工作
环境:python3.7,1080ti,tensorflow2.5(网上租的环境😂😂)
由于电脑问题,现在在jupytr上跑了,所以代码风格发生变化。
1.设置GPU或者cpu
设置cpu(电脑gpu跑不动就用这个将就一下)
from tensorflow import keras
from tensorflow.keras import layers, models
import os, PIL, pathlib
import matplotlib.pyplot as plt
import tensorflow as tf
import numpy as np
import tensorflow as tf
import tensorflow as tf
import os,PIL
# os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
#os.environ['CUDA_VISIBLE_DEVICES']='0'
os.environ['CUDA_VISIBLE_DEVICES']='2'
# os.environ['TF_CPP_MIN_LOG_LEVEL']='2'#屏蔽通知和警告信息
import os,PIL
os.environ['TF_XLA_FLAGS'] = '--tf_xla_enable_xla_devices'
from tensorflow import keras
keras.backend.clear_session()
设置gpu
from tensorflow import keras
from tensorflow.keras import layers,models
import os, PIL, pathlib
import matplotlib.pyplot as plt
import tensorflow as tf
import numpy as np
gpus = tf.config.list_physical_devices("GPU")
if gpus:
gpu0 = gpus[0] #如果有多个GPU,仅使用第0个GPU
tf.config.experimental.set_memory_growth(gpu0, True) #设置GPU显存用量按需使用
tf.config.set_visible_devices([gpu0],"GPU")
gpus
2.导入数据
data_dir = "./48-data/"
data_dir = pathlib.Path(data_dir)
3.查看数据
image_count = len(list(data_dir.glob('*/*.jpg')))
print("图片总数为:",image_count)
显示图片
roses = list(data_dir.glob('Jennifer Lawrence/*.jpg'))
PIL.Image.open(str(roses[0]))
二 数据预处理
1.加载数据
设置数据尺寸
batch_size = 32
img_height = 224
img_width = 224
设置dataset
train_ds = tf.keras.preprocessing.image_dataset_from_directory(
data_dir,
validation_split=0.1,
subset="training",
label_mode = "categorical",
seed=123,
image_size=(img_height, img_width),
batch_size=batch_size)
val_ds = tf.keras.preprocessing.image_dataset_from_directory(
data_dir,
validation_split=0.1,
subset="validation",
label_mode = "categorical",
seed=123,
image_size=(img_height, img_width),
batch_size=batch_size)
关于image_dataset_from_directory()的详细介绍可以参考文章:https://mtyjkh.blog.csdn.net/article/details/117018789
输出经过image_dataset_from_directory()分类后的标签
class_names = train_ds.class_names
print(class_names)
2.可视化数据
打印部分图片
plt.figure(figsize=(20, 10))
for images, labels in train_ds.take(1):
for i in range(20):
ax = plt.subplot(5, 10, i + 1)
plt.imshow(images[i].numpy().astype("uint8"))
plt.title(class_names[np.argmax(labels[i])])
plt.axis("off")
3.再次检查数据
输出数据的尺寸
for image_batch, labels_batch in train_ds:
print(image_batch.shape)
print(labels_batch.shape)
break
32:通道数 224x224为尺寸 3为rgb彩色通道(灰色为1)
32 :对应上面的32 17为标签种类的个数
4.配置数据集
shuffle:打乱数据集
prefetch:加速处理
AUTOTUNE = tf.data.AUTOTUNE
train_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)
三 搭建网络
本文神经网络为官方vgg16模型,我们需要做的是对最后一层按我们的类别进行分类即可。
调用官方
model = keras.applications.VGG16(include_top=False,weights=None)
global_average_layer = tf.keras.layers.GlobalAveragePooling2D()
prediction_layer = tf.keras.layers.Dense(len(class_names),activation='softmax')
model = tf.keras.Sequential([
model,
global_average_layer,
prediction_layer
])
关于VGG16函数参考下面文章
(8条消息) keras 自带VGG16 net 参数分析_vola9527的博客-CSDN博客
自己搭建
model = models.Sequential([
layers.experimental.preprocessing.Rescaling(1. / 255, input_shape=(img_height, img_width, 3)),
layers.Conv2D(filters=64, kernel_size=(3, 3), padding='same'), # 卷积层1
layers.BatchNormalization(), # BN层1
layers.Activation('relu'), # 激活层1
layers.Conv2D(filters=64, kernel_size=(3, 3), padding='same', ),
layers.BatchNormalization(), # BN层1
layers.Activation('relu') , # 激活层1
layers.MaxPool2D(pool_size=(2, 2), strides=2, padding='same'),
layers.Dropout(0.2), # dropout层
#
layers.Conv2D(filters=128, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization(), # BN层1
layers.Activation('relu'), # 激活层1
layers.Conv2D(filters=128, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization(), # BN层1
layers.Activation('relu'), # 激活层1
layers.MaxPool2D(pool_size=(2, 2), strides=2, padding='same'),
layers.Dropout(0.2), # dropout层
#
layers.Conv2D(filters=256, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization() , # BN层1
layers.Activation('relu'), # 激活层1
layers.Conv2D(filters=256, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization() , # BN层1
layers.Activation('relu') , # 激活层1
layers.Conv2D(filters=256, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization(),
layers.Activation('relu'),
layers.MaxPool2D(pool_size=(2, 2), strides=2, padding='same'),
layers.Dropout(0.2),
#
layers.Conv2D(filters=512, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization() , # BN层1
layers.Activation('relu') , # 激活层1
layers.Conv2D(filters=512, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization() , # BN层1
layers.Activation('relu'), # 激活层1
layers.Conv2D(filters=512, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization(),
layers.Activation('relu'),
layers.MaxPool2D(pool_size=(2, 2), strides=2, padding='same'),
layers.Dropout(0.2),
#
layers.Conv2D(filters=512, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization() , # BN层1
layers.Activation('relu'), # 激活层1
layers.Conv2D(filters=512, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization(), # BN层1
layers.Activation('relu'), # 激活层1
layers.Conv2D(filters=512, kernel_size=(3, 3), padding='same'),
layers.BatchNormalization(),
layers.Activation('relu'),
layers.MaxPool2D(pool_size=(2, 2), strides=2, padding='same'),
layers.Dropout(0.2),
#
# self.flatten = Flatten()
# self.f1 = Dense(512, activation='relu')
# self.d6 = Dropout(0.2)
# self.f2 = Dense(512, activation='relu')
# self.d7 = Dropout(0.2)
# self.f3 = Dense(10, activation='softmax')
layers.Flatten(), # Flatten层,连接卷积层与全连接层
layers.Dense(128, activation='relu'), # 全连接层,特征进一步提取
layers.Dense(len(class_names)) # 输出层,输出预期结果
])
四 训练模型
模型训练时,需要完成如下设置
损失函数(loss):衡量模型准确率
优化器(optimizer):根据损失函数进行优化更新
指标(metrics):监控训练过程,保存最优模型
1.设置动态学习率
# 设置初始学习率
initial_learning_rate = 1e-4
lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(
initial_learning_rate,
decay_steps=60, # 敲黑板!!!这里是指 steps,不是指epochs
decay_rate=0.96, # lr经过一次衰减就会变成 decay_rate*lr
staircase=True)
# 将指数衰减学习率送入优化器
optimizer = tf.keras.optimizers.Adam(learning_rate=lr_schedule)
model.compile(optimizer=optimizer,
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
这里的loss是设置为这个是因为:在dataset设置中
label_mode = "categorical",loss则为CategoricalCrossentropy
对于损失函数可以参考下面文章
(8条消息) tensorflow损失函数详解_重邮研究森的博客-CSDN博客
2.早期与保存最佳模型参数
关于ModelCheckpoint参考下面文章(8条消息) ModelCheckpoint 讲解【TensorFlow2入门手册】_K同学啊的博客-CSDN博客_modelcheckpoint函数
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping
epochs = 100
# 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)
3.模型训练
history = model.fit(train_ds,
validation_data=val_ds,
epochs=epochs,
callbacks=[checkpointer, earlystopper])
五 模型评估
1.Loss和Accuracy图
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs_range = range(len(loss))
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()
2.指定图片进行预测
# 加载效果最好的模型权重
model.load_weights('best_model.h5')
from PIL import Image
import numpy as np
img = Image.open("./48-data/Jennifer Lawrence/003_963a3627.jpg") #这里选择你需要预测的图片
img=np.array(img)
image = tf.image.resize(img, [img_height, img_width])
img_array = tf.expand_dims(image, 0)
predictions = model.predict(img_array) # 这里选用你已经训练好的模型
print("预测结果为:",class_names[np.argmax(predictions)])
3.总结
1.设置官方vgg16的weights=imagenet:准确率达到74,weights=None,准确率50-60
2.自搭建vgg16,最大池化:准确率40,平均池化:准确率57
3.损失函数为CategoricalCrossentropy比SparseCategoricalCrossentropy效果好