T6打卡—学习笔记

• 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
• 🍖 原作者：K同学啊

# 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

[]

# 导入并查看数据

data_dir = r"C:\Users\11054\Desktop\kLearning\t6_learning\data"
data_dir = pathlib.Path(data_dir)
image_count = len(list(data_dir.glob('*/*.jpg')))
print("图片总数为：",image_count)

图片总数为： 3600

roses = list(data_dir.glob('Jennifer Lawrence/*.jpg'))
PIL.Image.open(str(roses[0]))

# 数据预处理

batch_size = 32
img_height = 256
img_width = 256
"""
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
"""
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)

"""
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
"""
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)

Found 3600 files belonging to 17 classes.
Using 3240 files for training.
Found 3600 files belonging to 17 classes.
Using 360 files for validation.

# 查看标签和图片

class_names = train_ds.class_names
print(class_names)
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")

['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']

# 检查数据

for image_batch, labels_batch in train_ds:
    print(image_batch.shape)
    print(labels_batch.shape)
    break

(32, 224, 224, 3)
(32, 17)

# 配置数据集

AUTOTUNE = tf.data.AUTOTUNE

train_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)

# 构建CNN网络

"""
关于卷积核的计算不懂的可以参考文章：https://blog.csdn.net/qq_38251616/article/details/114278995

layers.Dropout(0.4) 作用是防止过拟合，提高模型的泛化能力。
关于Dropout层的更多介绍可以参考文章：https://mtyjkh.blog.csdn.net/article/details/115826689
"""

model = models.Sequential([
    layers.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.BatchNormalization(),                     # 添加BN层
    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.BatchNormalization(),                     # 添加BN层
    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.BatchNormalization(),                     # 添加BN层
    layers.AveragePooling2D((2, 2)),
    layers.Dropout(0.5),
    layers.Conv2D(128, (3, 3), activation='relu'),   # 卷积层4，卷积核3*3
    layers.BatchNormalization(),                     # 添加BN层
    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.BatchNormalization(),                     # 添加BN层
    layers.Dense(len(class_names))                   # 输出层，输出预期结果
])

model.summary()  # 打印网络结构

C:\Users\11054\.conda\envs\py311\Lib\site-packages\keras\src\layers\preprocessing\tf_data_layer.py:18: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(**kwargs)
C:\Users\11054\.conda\envs\py311\Lib\site-packages\keras\src\layers\convolutional\base_conv.py:107: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(activity_regularizer=activity_regularizer, **kwargs)

Model: "sequential_2"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type)                         ┃ Output Shape                ┃         Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ rescaling_2 (Rescaling)              │ (None, 224, 224, 3)         │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_8 (Conv2D)                    │ (None, 222, 222, 16)        │             448 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ batch_normalization                  │ (None, 222, 222, 16)        │              64 │
│ (BatchNormalization)                 │                             │                 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ average_pooling2d_6                  │ (None, 111, 111, 16)        │               0 │
│ (AveragePooling2D)                   │                             │                 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_9 (Conv2D)                    │ (None, 109, 109, 32)        │           4,640 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ batch_normalization_1                │ (None, 109, 109, 32)        │             128 │
│ (BatchNormalization)                 │                             │                 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ average_pooling2d_7                  │ (None, 54, 54, 32)          │               0 │
│ (AveragePooling2D)                   │                             │                 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dropout_6 (Dropout)                  │ (None, 54, 54, 32)          │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_10 (Conv2D)                   │ (None, 52, 52, 64)          │          18,496 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ batch_normalization_2                │ (None, 52, 52, 64)          │             256 │
│ (BatchNormalization)                 │                             │                 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ average_pooling2d_8                  │ (None, 26, 26, 64)          │               0 │
│ (AveragePooling2D)                   │                             │                 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dropout_7 (Dropout)                  │ (None, 26, 26, 64)          │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ conv2d_11 (Conv2D)                   │ (None, 24, 24, 128)         │          73,856 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ batch_normalization_3                │ (None, 24, 24, 128)         │             512 │
│ (BatchNormalization)                 │                             │                 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dropout_8 (Dropout)                  │ (None, 24, 24, 128)         │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ flatten_2 (Flatten)                  │ (None, 73728)               │               0 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_4 (Dense)                      │ (None, 128)                 │       9,437,312 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ batch_normalization_4                │ (None, 128)                 │             512 │
│ (BatchNormalization)                 │                             │                 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ dense_5 (Dense)                      │ (None, 17)                  │           2,193 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘

 Total params: 9,538,417 (36.39 MB)

 Trainable params: 9,537,681 (36.38 MB)

 Non-trainable params: 736 (2.88 KB)

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'])

损失函数Loss详解：

1. binary_crossentropy（对数损失函数）

与 sigmoid 相对应的损失函数，针对于二分类问题。

2. categorical_crossentropy（多分类的对数损失函数）

与 softmax 相对应的损失函数，如果是one-hot编码，则使用 categorical_crossentropy

• 调用方法一：

model.compile(optimizer=“adam”,
loss=‘categorical_crossentropy’,
metrics=[‘accuracy’])

• 调用方法二：

model.compile(optimizer=“adam”,
loss=tf.keras.losses.CategoricalCrossentropy(),
metrics=[‘accuracy’])

3. sparse_categorical_crossentropy（稀疏性多分类的对数损失函数）

与 softmax 相对应的损失函数，如果是整数编码，则使用 sparse_categorical_crossentropy

• 调用方法一：

model.compile(optimizer=“adam”,
loss=‘sparse_categorical_crossentropy’,
metrics=[‘accuracy’])

• 调用方法二：

model.compile(optimizer=“adam”,
loss=tf.keras.losses.SparseCategoricalCrossentropy(),
metrics=[‘accuracy’])

函数原型

tf.keras.losses.SparseCategoricalCrossentropy(
from_logits=False,
reduction=losses_utils.ReductionV2.AUTO,
name=‘sparse_categorical_crossentropy’
)

参数说明：

from_logits: 为True时，会将y_pred转化为概率（用softmax）,否则不进行转换，通常情况下用True结果更稳定；
reduction：类型为tf.keras.losses.Reduction，对loss进行处理，默认是AUTO；
name: name

设置早停

EarlyStopping()参数说明：

• monitor: 被监测的数据。
• min_delta: 在被监测的数据中被认为是提升的最小变化， 例如，小于 min_delta 的绝对变化会被认为没有提升。
• patience: 没有进步的训练轮数，在这之后训练就会被停止。
• verbose: 详细信息模式。
• mode: {auto, min, max} 其中之一。 在 min 模式中， 当被监测的数据停止下降，训练就会停止；在 max 模式中，当被监测的数据停止上升，训练就会停止；在 auto 模式中，方向会自动从被监测的数据的名字中判断出来。
• baseline: 要监控的数量的基准值。 如果模型没有显示基准的改善，训练将停止。
• estore_best_weights: 是否从具有监测数量的最佳值的时期恢复模型权重。 如果为 False，则使用在训练的最后一步获得的模型权重。

from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping

epochs = 100

# 保存最佳模型参数
checkpointer = ModelCheckpoint('best_model.weights.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)

训练模型

history = model.fit(train_ds,
                    validation_data=val_ds,
                    epochs=epochs,
                    callbacks=[checkpointer, earlystopper])

Epoch 1/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 408ms/step - accuracy: 0.1376 - loss: 3.1025
Epoch 1: val_accuracy improved from -inf to 0.08056, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m46s[0m 419ms/step - accuracy: 0.1379 - loss: 3.1002 - val_accuracy: 0.0806 - val_loss: 2.8348
Epoch 2/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 406ms/step - accuracy: 0.3448 - loss: 2.0340
Epoch 2: val_accuracy improved from 0.08056 to 0.11111, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 416ms/step - accuracy: 0.3450 - loss: 2.0337 - val_accuracy: 0.1111 - val_loss: 3.2089
Epoch 3/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 408ms/step - accuracy: 0.4646 - loss: 1.6663
Epoch 3: val_accuracy improved from 0.11111 to 0.16389, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m43s[0m 418ms/step - accuracy: 0.4647 - loss: 1.6662 - val_accuracy: 0.1639 - val_loss: 2.9885
Epoch 4/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 402ms/step - accuracy: 0.5494 - loss: 1.4220
Epoch 4: val_accuracy improved from 0.16389 to 0.21667, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 412ms/step - accuracy: 0.5496 - loss: 1.4217 - val_accuracy: 0.2167 - val_loss: 2.8192
Epoch 5/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 401ms/step - accuracy: 0.6510 - loss: 1.1654
Epoch 5: val_accuracy improved from 0.21667 to 0.30833, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 410ms/step - accuracy: 0.6509 - loss: 1.1655 - val_accuracy: 0.3083 - val_loss: 2.4227
Epoch 6/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 413ms/step - accuracy: 0.7351 - loss: 0.9649
Epoch 6: val_accuracy improved from 0.30833 to 0.40000, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m43s[0m 423ms/step - accuracy: 0.7349 - loss: 0.9651 - val_accuracy: 0.4000 - val_loss: 1.9446
Epoch 7/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 403ms/step - accuracy: 0.7599 - loss: 0.8986
Epoch 7: val_accuracy improved from 0.40000 to 0.45000, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 413ms/step - accuracy: 0.7599 - loss: 0.8984 - val_accuracy: 0.4500 - val_loss: 1.8153
Epoch 8/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 403ms/step - accuracy: 0.8122 - loss: 0.7271
Epoch 8: val_accuracy improved from 0.45000 to 0.45833, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 413ms/step - accuracy: 0.8121 - loss: 0.7272 - val_accuracy: 0.4583 - val_loss: 1.7950
Epoch 9/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 402ms/step - accuracy: 0.8508 - loss: 0.6247
Epoch 9: val_accuracy improved from 0.45833 to 0.47778, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 411ms/step - accuracy: 0.8507 - loss: 0.6248 - val_accuracy: 0.4778 - val_loss: 1.7466
Epoch 10/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 410ms/step - accuracy: 0.8809 - loss: 0.5505
Epoch 10: val_accuracy improved from 0.47778 to 0.48333, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m43s[0m 419ms/step - accuracy: 0.8809 - loss: 0.5506 - val_accuracy: 0.4833 - val_loss: 1.6903
Epoch 11/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 414ms/step - accuracy: 0.9110 - loss: 0.4850
Epoch 11: val_accuracy improved from 0.48333 to 0.48889, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m43s[0m 423ms/step - accuracy: 0.9109 - loss: 0.4850 - val_accuracy: 0.4889 - val_loss: 1.6706
Epoch 12/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 411ms/step - accuracy: 0.9331 - loss: 0.4198
Epoch 12: val_accuracy improved from 0.48889 to 0.51389, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m43s[0m 421ms/step - accuracy: 0.9330 - loss: 0.4199 - val_accuracy: 0.5139 - val_loss: 1.6316
Epoch 13/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 421ms/step - accuracy: 0.9469 - loss: 0.3716
Epoch 13: val_accuracy did not improve from 0.51389
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m44s[0m 430ms/step - accuracy: 0.9469 - loss: 0.3717 - val_accuracy: 0.5000 - val_loss: 1.6615
Epoch 14/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 407ms/step - accuracy: 0.9690 - loss: 0.3244
Epoch 14: val_accuracy improved from 0.51389 to 0.52222, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m43s[0m 418ms/step - accuracy: 0.9689 - loss: 0.3245 - val_accuracy: 0.5222 - val_loss: 1.6336
Epoch 15/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 403ms/step - accuracy: 0.9631 - loss: 0.3060
Epoch 15: val_accuracy did not improve from 0.52222
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 411ms/step - accuracy: 0.9631 - loss: 0.3060 - val_accuracy: 0.5056 - val_loss: 1.6253
Epoch 16/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 404ms/step - accuracy: 0.9661 - loss: 0.2795
Epoch 16: val_accuracy improved from 0.52222 to 0.53333, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 414ms/step - accuracy: 0.9661 - loss: 0.2796 - val_accuracy: 0.5333 - val_loss: 1.6280
Epoch 17/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 406ms/step - accuracy: 0.9797 - loss: 0.2492
Epoch 17: val_accuracy did not improve from 0.53333
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 414ms/step - accuracy: 0.9797 - loss: 0.2493 - val_accuracy: 0.5278 - val_loss: 1.6102
Epoch 18/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 402ms/step - accuracy: 0.9768 - loss: 0.2485
Epoch 18: val_accuracy improved from 0.53333 to 0.53889, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 412ms/step - accuracy: 0.9768 - loss: 0.2485 - val_accuracy: 0.5389 - val_loss: 1.5958
Epoch 19/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 402ms/step - accuracy: 0.9843 - loss: 0.2263
Epoch 19: val_accuracy did not improve from 0.53889
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 410ms/step - accuracy: 0.9843 - loss: 0.2263 - val_accuracy: 0.5333 - val_loss: 1.5964
Epoch 20/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 402ms/step - accuracy: 0.9820 - loss: 0.2071
Epoch 20: val_accuracy did not improve from 0.53889
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 411ms/step - accuracy: 0.9820 - loss: 0.2071 - val_accuracy: 0.5222 - val_loss: 1.6034
Epoch 21/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 401ms/step - accuracy: 0.9856 - loss: 0.1909
Epoch 21: val_accuracy did not improve from 0.53889
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 410ms/step - accuracy: 0.9856 - loss: 0.1910 - val_accuracy: 0.5389 - val_loss: 1.6007
Epoch 22/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 403ms/step - accuracy: 0.9885 - loss: 0.1841
Epoch 22: val_accuracy did not improve from 0.53889
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 411ms/step - accuracy: 0.9885 - loss: 0.1841 - val_accuracy: 0.5139 - val_loss: 1.6120
Epoch 23/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 405ms/step - accuracy: 0.9905 - loss: 0.1742
Epoch 23: val_accuracy did not improve from 0.53889
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 413ms/step - accuracy: 0.9905 - loss: 0.1742 - val_accuracy: 0.5333 - val_loss: 1.5867
Epoch 24/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 404ms/step - accuracy: 0.9925 - loss: 0.1639
Epoch 24: val_accuracy did not improve from 0.53889
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 412ms/step - accuracy: 0.9925 - loss: 0.1640 - val_accuracy: 0.5361 - val_loss: 1.5717
Epoch 25/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 403ms/step - accuracy: 0.9908 - loss: 0.1616
Epoch 25: val_accuracy did not improve from 0.53889
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 411ms/step - accuracy: 0.9908 - loss: 0.1616 - val_accuracy: 0.5389 - val_loss: 1.5785
Epoch 26/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 403ms/step - accuracy: 0.9889 - loss: 0.1520
Epoch 26: val_accuracy did not improve from 0.53889
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 412ms/step - accuracy: 0.9890 - loss: 0.1520 - val_accuracy: 0.5306 - val_loss: 1.5847
Epoch 27/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 405ms/step - accuracy: 0.9947 - loss: 0.1522
Epoch 27: val_accuracy improved from 0.53889 to 0.54167, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 416ms/step - accuracy: 0.9947 - loss: 0.1522 - val_accuracy: 0.5417 - val_loss: 1.5715
Epoch 28/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 406ms/step - accuracy: 0.9933 - loss: 0.1477
Epoch 28: val_accuracy did not improve from 0.54167
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 415ms/step - accuracy: 0.9933 - loss: 0.1476 - val_accuracy: 0.5361 - val_loss: 1.5867
Epoch 29/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 406ms/step - accuracy: 0.9964 - loss: 0.1392
Epoch 29: val_accuracy did not improve from 0.54167
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 415ms/step - accuracy: 0.9964 - loss: 0.1392 - val_accuracy: 0.5361 - val_loss: 1.5947
Epoch 30/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 406ms/step - accuracy: 0.9953 - loss: 0.1342
Epoch 30: val_accuracy improved from 0.54167 to 0.55556, saving model to best_model.weights.h5
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 416ms/step - accuracy: 0.9953 - loss: 0.1342 - val_accuracy: 0.5556 - val_loss: 1.5768
Epoch 31/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 406ms/step - accuracy: 0.9948 - loss: 0.1326
Epoch 31: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 415ms/step - accuracy: 0.9949 - loss: 0.1325 - val_accuracy: 0.5472 - val_loss: 1.5713
Epoch 32/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 402ms/step - accuracy: 0.9962 - loss: 0.1262
Epoch 32: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 411ms/step - accuracy: 0.9962 - loss: 0.1262 - val_accuracy: 0.5472 - val_loss: 1.5751
Epoch 33/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 405ms/step - accuracy: 0.9975 - loss: 0.1147
Epoch 33: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 414ms/step - accuracy: 0.9975 - loss: 0.1148 - val_accuracy: 0.5417 - val_loss: 1.5769
Epoch 34/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 404ms/step - accuracy: 0.9973 - loss: 0.1205
Epoch 34: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 414ms/step - accuracy: 0.9973 - loss: 0.1205 - val_accuracy: 0.5528 - val_loss: 1.5657
Epoch 35/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 406ms/step - accuracy: 0.9974 - loss: 0.1157
Epoch 35: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 414ms/step - accuracy: 0.9974 - loss: 0.1158 - val_accuracy: 0.5500 - val_loss: 1.5527
Epoch 36/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 401ms/step - accuracy: 0.9970 - loss: 0.1102
Epoch 36: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 409ms/step - accuracy: 0.9970 - loss: 0.1102 - val_accuracy: 0.5472 - val_loss: 1.5561
Epoch 37/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 399ms/step - accuracy: 0.9978 - loss: 0.1072
Epoch 37: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 407ms/step - accuracy: 0.9978 - loss: 0.1072 - val_accuracy: 0.5417 - val_loss: 1.5641
Epoch 38/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 400ms/step - accuracy: 0.9969 - loss: 0.1066
Epoch 38: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 408ms/step - accuracy: 0.9969 - loss: 0.1066 - val_accuracy: 0.5556 - val_loss: 1.5585
Epoch 39/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 399ms/step - accuracy: 0.9986 - loss: 0.1058
Epoch 39: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 408ms/step - accuracy: 0.9986 - loss: 0.1058 - val_accuracy: 0.5472 - val_loss: 1.5690
Epoch 40/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 399ms/step - accuracy: 0.9974 - loss: 0.1040
Epoch 40: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 408ms/step - accuracy: 0.9974 - loss: 0.1040 - val_accuracy: 0.5472 - val_loss: 1.5677
Epoch 41/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 401ms/step - accuracy: 0.9966 - loss: 0.0998
Epoch 41: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 409ms/step - accuracy: 0.9966 - loss: 0.0998 - val_accuracy: 0.5528 - val_loss: 1.5675
Epoch 42/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 402ms/step - accuracy: 0.9980 - loss: 0.0941
Epoch 42: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 410ms/step - accuracy: 0.9980 - loss: 0.0941 - val_accuracy: 0.5500 - val_loss: 1.5738
Epoch 43/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 400ms/step - accuracy: 0.9968 - loss: 0.1026
Epoch 43: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 408ms/step - accuracy: 0.9969 - loss: 0.1026 - val_accuracy: 0.5472 - val_loss: 1.5729
Epoch 44/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 404ms/step - accuracy: 0.9997 - loss: 0.0937
Epoch 44: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 412ms/step - accuracy: 0.9997 - loss: 0.0937 - val_accuracy: 0.5472 - val_loss: 1.5712
Epoch 45/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 404ms/step - accuracy: 0.9960 - loss: 0.0987
Epoch 45: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 413ms/step - accuracy: 0.9960 - loss: 0.0986 - val_accuracy: 0.5444 - val_loss: 1.5635
Epoch 46/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 402ms/step - accuracy: 0.9978 - loss: 0.0934
Epoch 46: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 410ms/step - accuracy: 0.9978 - loss: 0.0934 - val_accuracy: 0.5444 - val_loss: 1.5577
Epoch 47/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 402ms/step - accuracy: 0.9983 - loss: 0.0929
Epoch 47: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 411ms/step - accuracy: 0.9983 - loss: 0.0929 - val_accuracy: 0.5444 - val_loss: 1.5616
Epoch 48/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 403ms/step - accuracy: 0.9989 - loss: 0.0870
Epoch 48: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 411ms/step - accuracy: 0.9989 - loss: 0.0870 - val_accuracy: 0.5472 - val_loss: 1.5651
Epoch 49/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 405ms/step - accuracy: 0.9995 - loss: 0.0908
Epoch 49: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 414ms/step - accuracy: 0.9995 - loss: 0.0909 - val_accuracy: 0.5417 - val_loss: 1.5546
Epoch 50/100
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 401ms/step - accuracy: 0.9980 - loss: 0.0859
Epoch 50: val_accuracy did not improve from 0.55556
[1m102/102[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m42s[0m 409ms/step - accuracy: 0.9980 - loss: 0.0859 - val_accuracy: 0.5500 - val_loss: 1.5478
Epoch 50: early stopping

评估模型

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()

使用模型预测

# 加载效果最好的模型权重
model.load_weights('best_model.weights.h5')
from PIL import Image
import numpy as np

img = Image.open(r"C:\Users\11054\Desktop\kLearning\t6_learning\data\Angelina Jolie\001_fe3347c0.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)])

[1m1/1[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 85ms/step
预测结果为： Angelina Jolie

个人总结

jupyter nbconvert --to markdown file_path.ipynb
模型训练精度效果良好但是测试精度较差
添加BN层后测试精度得到提高