- 🍨 本文为🔗365天深度学习训练营 内部限免文章(版权归 K同学啊 所有)
- 🍦 参考文章地址: 🔗深度学习实战训练 | 第6周:好莱坞明星识别
- 🍖 作者:K同学啊 | 接辅导、程序定制
🏡 运行环境:
电脑系统:Windows 10
语言环境:python 3.10
编译器:Pycharm 2022.1.1
深度学习环境:Pytorch
3.检查数据
一、前期准备
1.设置GPU
根据自己的电脑配置进行设置,如果使用的是CPU可以忽略。
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.查看和输出数据
image_count = len(list(data_dir.glob('*/*.jpg')))
print("图片总数为:",image_count)
roses = list(data_dir.glob('Jennifer Lawrence/*.jpg'))
PIL.Image.open(str(roses[0]))
Out:
图片总数为: 1800
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
测试集与验证集的关系:
- 验证集并没有参与训练过程梯度下降过程的,狭义上来讲是没有参与模型的参数训练更新的。
- 但是广义上来讲,验证集存在的意义确实参与了一个“人工调参”的过程,我们根据每一个epoch训练之后模型在valid data上的表现来决定是否需要训练进行early stop,或者根据这个过程模型的性能变化来调整模型的超参数,如学习率,batch_size等等。
- 因此,我们也可以认为,验证集也参与了训练,但是并没有使得模型去overfit验证集
batch_size = 32
img_height = 224
img_width = 224
label_mode:
int
:标签将被编码成整数(使用的损失函数应为:sparse_categorical_crossentropy loss)。categorical
:标签将被编码为分类向量(使用的损失函数应为:categorical_crossentropy loss)。
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)
Out:
Found 1800 files belonging to 17 classes. Using 1620 files for training.
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)
Out:
Found 1800 files belonging to 17 classes. Using 180 files for validation.
可以通过class_names输出数据集的标签。标签将按字母顺序对应于目录名称。
class_names = train_ds.class_names
print(class_names)
Out:
['Angelina Jolie', 'Brad Pitt', 'Denzel Washington', 'Hugh Jackman', 'Jennifer Lawrence', 'Johnny Depp', 'Kate Winslet', 'Leonardo DiCaprio', 'Megan Fox', 'Natalie Portman', 'Nicole Kidman', 'Robert Downey Jr', 'Sandra Bullock', 'Scarlett Johansson', 'Tom Cruise', 'Tom Hanks', 'Will Smith']
2.可视化数据
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
for image_batch, labels_batch in train_ds:
print(image_batch.shape)
print(labels_batch.shape)
break
Out:
(32, 224, 224, 3) (32, 17)
Image_batch
是形状的张量(32,224,224,3)。这是一批形状224x224x3的32张图片(最后一维指的是彩色通道RGB)。Label_batch
是形状(32,)的张量,这些标签对应32张图片
4.配置数据集
- shuffle() :打乱数据,关于此函数的详细介绍可以参考:https://zhuanlan.zhihu.com/p/42417456
- prefetch() :预取数据,加速运行
prefetch()
功能详细介绍:CPU 正在准备数据时,加速器处于空闲状态。相反,当加速器正在训练模型时,CPU 处于空闲状态。因此,训练所用的时间是 CPU 预处理时间和加速器训练时间的总和。prefetch()
将训练步骤的预处理和模型执行过程重叠到一起。当加速器正在执行第 N 个训练步时,CPU 正在准备第 N+1 步的数据。这样做不仅可以最大限度地缩短训练的单步用时(而不是总用时),而且可以缩短提取和转换数据所需的时间。如果不使用prefetch()
,CPU 和 GPU/TPU 在大部分时间都处于空闲状态:
使用prefetch()
可显著减少空闲时间:
- cache() :将数据集缓存到内存当中,加速运行
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、构建CNN网络
卷积神经网络(CNN)的输入是张量 (Tensor) 形式的 (image_height, image_width, color_channels)
,包含了图像高度、宽度及颜色信息。不需要输入batch size
。color_channels 为 (R,G,B) 分别对应 RGB 的三个颜色通道(color channel)。在此示例中,我们的 CNN 输入的形状是 (224, 224, 4)
即彩色图像。我们需要在声明第一层时将形状赋值给参数input_shape
。
model = models.Sequential([
layers.experimental.preprocessing.Rescaling(1./255, input_shape=(img_height, img_width, 3)),
layers.Conv2D(16, (3, 3), activation='relu', input_shape=(img_height, img_width, 3)), # 卷积层1,卷积核3*3
layers.AveragePooling2D((2, 2)), # 池化层1,2*2采样
layers.Conv2D(32, (3, 3), activation='relu'), # 卷积层2,卷积核3*3
layers.AveragePooling2D((2, 2)), # 池化层2,2*2采样
layers.Dropout(0.5),
layers.Conv2D(64, (3, 3), activation='relu'), # 卷积层3,卷积核3*3
layers.AveragePooling2D((2, 2)),
layers.Dropout(0.5),
layers.Conv2D(128, (3, 3), activation='relu'), # 卷积层3,卷积核3*3
layers.Dropout(0.5),
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
])
model.summary() # 打印网络结构
Model: "sequential" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= rescaling (Rescaling) (None, 224, 224, 3) 0 _________________________________________________________________ conv2d (Conv2D) (None, 222, 222, 16) 448 _________________________________________________________________ average_pooling2d (AveragePo (None, 111, 111, 16) 0 _________________________________________________________________ conv2d_1 (Conv2D) (None, 109, 109, 32) 4640 _________________________________________________________________ average_pooling2d_1 (Average (None, 54, 54, 32) 0 _________________________________________________________________ dropout (Dropout) (None, 54, 54, 32) 0 _________________________________________________________________ conv2d_2 (Conv2D) (None, 52, 52, 64) 18496 _________________________________________________________________ average_pooling2d_2 (Average (None, 26, 26, 64) 0 _________________________________________________________________ dropout_1 (Dropout) (None, 26, 26, 64) 0 _________________________________________________________________ conv2d_3 (Conv2D) (None, 24, 24, 128) 73856 _________________________________________________________________ dropout_2 (Dropout) (None, 24, 24, 128) 0 _________________________________________________________________ flatten (Flatten) (None, 73728) 0 _________________________________________________________________ dense (Dense) (None, 128) 9437312 _________________________________________________________________ dense_1 (Dense) (None, 17) 2193 ================================================================= Total params: 9,536,945 Trainable params: 9,536,945 Non-trainable params: 0 _________________________________________________________________
四、训练模型
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。
- 指标(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。
# 将指数衰减学习率送入优化器
optimizer = tf.keras.optimizers.Adam()
model.compile(optimizer=optimizer,
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
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)
history = model.fit(train_ds,
validation_data=val_ds,
epochs=50)
Epoch 1/50 51/51 [==============================] - 5s 20ms/step - loss: 2.8329 - accuracy: 0.0895 - val_loss: 2.7887 - val_accuracy: 0.1389 Epoch 2/50 51/51 [==============================] - 1s 12ms/step - loss: 2.7283 - accuracy: 0.1296 - val_loss: 2.6685 - val_accuracy: 0.1889 Epoch 3/50 51/51 [==============================] - 1s 12ms/step - loss: 2.5299 - accuracy: 0.1914 - val_loss: 2.5374 - val_accuracy: 0.1611 Epoch 4/50 51/51 [==============================] - 1s 12ms/step - loss: 2.3337 - accuracy: 0.2469 - val_loss: 2.4386 - val_accuracy: 0.2333 Epoch 5/50 51/51 [==============================] - 1s 12ms/step - loss: 2.1101 - accuracy: 0.3130 - val_loss: 2.3895 - val_accuracy: 0.2167 Epoch 6/50 51/51 [==============================] - 1s 12ms/step - loss: 1.8293 - accuracy: 0.4080 - val_loss: 2.2831 - val_accuracy: 0.2500 Epoch 7/50 51/51 [==============================] - 1s 12ms/step - loss: 1.4679 - accuracy: 0.5043 - val_loss: 2.3525 - val_accuracy: 0.2611 Epoch 8/50 51/51 [==============================] - 1s 12ms/step - loss: 1.0419 - accuracy: 0.6667 - val_loss: 2.6046 - val_accuracy: 0.2611 Epoch 9/50 51/51 [==============================] - 1s 12ms/step - loss: 0.7165 - accuracy: 0.7599 - val_loss: 2.9672 - val_accuracy: 0.2944 Epoch 10/50 51/51 [==============================] - 1s 12ms/step - loss: 0.4798 - accuracy: 0.8451 - val_loss: 3.3681 - val_accuracy: 0.3278 Epoch 11/50 51/51 [==============================] - 1s 12ms/step - loss: 0.3122 - accuracy: 0.8981 - val_loss: 4.1233 - val_accuracy: 0.3167 Epoch 12/50 51/51 [==============================] - 1s 12ms/step - loss: 0.2503 - accuracy: 0.9204 - val_loss: 4.4343 - val_accuracy: 0.3444 Epoch 13/50 51/51 [==============================] - 1s 12ms/step - loss: 0.2311 - accuracy: 0.9272 - val_loss: 4.2779 - val_accuracy: 0.3222 Epoch 14/50 51/51 [==============================] - 1s 12ms/step - loss: 0.1623 - accuracy: 0.9444 - val_loss: 4.6655 - val_accuracy: 0.3389 Epoch 15/50 51/51 [==============================] - 1s 12ms/step - loss: 0.1790 - accuracy: 0.9438 - val_loss: 4.6963 - val_accuracy: 0.3278 Epoch 16/50 51/51 [==============================] - 1s 12ms/step - loss: 0.1234 - accuracy: 0.9605 - val_loss: 5.3258 - val_accuracy: 0.3167 Epoch 17/50 51/51 [==============================] - 1s 12ms/step - loss: 0.1344 - accuracy: 0.9623 - val_loss: 5.1431 - val_accuracy: 0.3000 Epoch 18/50 51/51 [==============================] - 1s 12ms/step - loss: 0.1094 - accuracy: 0.9654 - val_loss: 5.7386 - val_accuracy: 0.3000 Epoch 19/50 51/51 [==============================] - 1s 12ms/step - loss: 0.1050 - accuracy: 0.9691 - val_loss: 5.4096 - val_accuracy: 0.3444 Epoch 20/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0791 - accuracy: 0.9784 - val_loss: 5.7964 - val_accuracy: 0.3389 Epoch 21/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0507 - accuracy: 0.9802 - val_loss: 6.0106 - val_accuracy: 0.3056 Epoch 22/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0691 - accuracy: 0.9821 - val_loss: 6.0377 - val_accuracy: 0.3222 Epoch 23/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0781 - accuracy: 0.9753 - val_loss: 5.2907 - val_accuracy: 0.3278 Epoch 24/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0638 - accuracy: 0.9753 - val_loss: 5.6900 - val_accuracy: 0.3500 Epoch 25/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0575 - accuracy: 0.9796 - val_loss: 5.6980 - val_accuracy: 0.2944 Epoch 26/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0595 - accuracy: 0.9827 - val_loss: 6.0583 - val_accuracy: 0.3111 Epoch 27/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0567 - accuracy: 0.9858 - val_loss: 6.1378 - val_accuracy: 0.3222 Epoch 28/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0506 - accuracy: 0.9846 - val_loss: 6.3385 - val_accuracy: 0.3167 Epoch 29/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0480 - accuracy: 0.9821 - val_loss: 6.7924 - val_accuracy: 0.3056 Epoch 30/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0511 - accuracy: 0.9809 - val_loss: 6.8792 - val_accuracy: 0.2944 Epoch 31/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0364 - accuracy: 0.9883 - val_loss: 7.1183 - val_accuracy: 0.3500 Epoch 32/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0635 - accuracy: 0.9802 - val_loss: 6.4413 - val_accuracy: 0.3111 Epoch 33/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0429 - accuracy: 0.9883 - val_loss: 6.2766 - val_accuracy: 0.3222 Epoch 34/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0494 - accuracy: 0.9833 - val_loss: 6.5672 - val_accuracy: 0.2889 Epoch 35/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0334 - accuracy: 0.9870 - val_loss: 7.1044 - val_accuracy: 0.3778 Epoch 36/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0367 - accuracy: 0.9877 - val_loss: 6.9240 - val_accuracy: 0.3389 Epoch 37/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0371 - accuracy: 0.9870 - val_loss: 6.8424 - val_accuracy: 0.3111 Epoch 38/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0251 - accuracy: 0.9914 - val_loss: 7.3906 - val_accuracy: 0.3222 Epoch 39/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0364 - accuracy: 0.9877 - val_loss: 7.3986 - val_accuracy: 0.3056 Epoch 40/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0302 - accuracy: 0.9889 - val_loss: 7.7252 - val_accuracy: 0.2833 Epoch 41/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0618 - accuracy: 0.9815 - val_loss: 6.4630 - val_accuracy: 0.2889 Epoch 42/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0448 - accuracy: 0.9827 - val_loss: 6.6955 - val_accuracy: 0.3000 Epoch 43/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0290 - accuracy: 0.9914 - val_loss: 6.4365 - val_accuracy: 0.3000 Epoch 44/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0245 - accuracy: 0.9926 - val_loss: 6.4801 - val_accuracy: 0.3000 Epoch 45/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0239 - accuracy: 0.9926 - val_loss: 6.9107 - val_accuracy: 0.3000 Epoch 46/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0307 - accuracy: 0.9907 - val_loss: 6.4642 - val_accuracy: 0.3000 Epoch 47/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0313 - accuracy: 0.9901 - val_loss: 7.5312 - val_accuracy: 0.2833 Epoch 48/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0284 - accuracy: 0.9926 - val_loss: 7.2625 - val_accuracy: 0.2889 Epoch 49/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0538 - accuracy: 0.9840 - val_loss: 6.8608 - val_accuracy: 0.2833 Epoch 50/50 51/51 [==============================] - 1s 12ms/step - loss: 0.0390
五、模型评估
1.Loss与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
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)])
预测结果为: Jennifer Lawrence