第T6周：好莱坞明星识别

• 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

要求：
1.使用categorical_crossentropy（多分类的对数损失函数）完成本次选题。
2.探究不同损失函数的使用场景与代码实现。

拔高（可选）：
1.自己搭建VGG-16网络框架
2.调用官方的VGG-16网络框架
3.使用VGG-16算法训练该模型

探索（难度有点大）
1.准确率达到60%

我的环境：
●操作系统：ubuntu 22.04
●语言环境：python 3.8.10
●编译器：jupyter notebook
●深度学习框架：tensorflow-gpu 2.9.0
●显卡（GPU）：RTX 3090(24GB) * 1
●数据集：好莱坞明星数据集

一、前期工作

1. 设置GPU(如果使用的是CPU可以忽略这步)

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

代码输出：

[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

2. 导入数据

data_dir = "./T6/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)

代码输出：

图片总数为： 1800

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

代码输出：

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：

1.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。
2.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。
3.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 = 256   #原代码是224
img_width = 256    #原代码是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）。

"""
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)

代码输出：

Found 1800 files belonging to 17 classes.
Using 1620 files for training.

"""
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)

代码输出：

Found 1800 files belonging to 17 classes.
Using 180 files for validation.

我们可以通过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)

代码输出：

['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(15):            #原代码是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")

代码输出：

3. 再次检查数据

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

代码输出：

(32, 256, 256, 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。

AUTOTUNE = tf.data.AUTOTUNE

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

5. 数据增强，并将其嵌入model中（原代码没有这部分的）

data_augmentation = tf.keras.Sequential([
  tf.keras.layers.experimental.preprocessing.RandomFlip("horizontal_and_vertical"),
  tf.keras.layers.experimental.preprocessing.RandomRotation(0.2),
])

model = tf.keras.Sequential([
  data_augmentation,
  layers.Conv2D(16, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
])

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, 3)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。

这里的(224, 224, 3)经过上面的代码修改，已经是（256, 256, 3)了。

网络结构图

原代码如下：

"""
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.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

下面的代码是我根据上面的代码（原代码）进行修改的，主要是添加Batch Normalization (BN) 和修改Dropout 层的参数：

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)),  
    layers.Conv2D(16, (3, 3), activation='relu'), 
    layers.BatchNormalization(), 
    layers.AveragePooling2D((2, 2)),             
    
    layers.Conv2D(32, (3, 3), activation='relu'),  
    layers.Conv2D(32, (3, 3), activation='relu'),  
    layers.BatchNormalization(), 
    layers.AveragePooling2D((2, 2)),              
    layers.Dropout(0.6),  
    
    layers.Conv2D(64, (3, 3), activation='relu'),  
    layers.Conv2D(64, (3, 3), activation='relu'),  
    layers.BatchNormalization(),  
    layers.AveragePooling2D((2, 2)),     
    layers.Dropout(0.6),
    
    layers.Conv2D(128, (3, 3), activation='relu'), 
    layers.Conv2D(128, (3, 3), activation='relu'),  
    layers.BatchNormalization(),  # 添加BN层4
    layers.Dropout(0.6), 
    
    
    layers.Conv2D(516, (3, 3), activation='relu'), 
    layers.Conv2D(516, (3, 3), activation='relu'),  
    layers.BatchNormalization(),  
    layers.Dropout(0.6),
    
    layers.Flatten(),                       
    layers.Dense(516, activation='relu'),  
    layers.BatchNormalization(), 
    layers.Dense(len(class_names))               
])

model.summary()  # 打印网络结构

代码输出：

Model: "sequential_11"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 rescaling_4 (Rescaling)     (None, 256, 256, 3)       0         
                                                                 
 conv2d_24 (Conv2D)          (None, 254, 254, 16)      448       
                                                                 
 conv2d_25 (Conv2D)          (None, 252, 252, 16)      2320      
                                                                 
 batch_normalization_22 (Bat  (None, 252, 252, 16)     64        
 chNormalization)                                                
                                                                 
 average_pooling2d_12 (Avera  (None, 126, 126, 16)     0         
 gePooling2D)                                                    
                                                                 
 conv2d_26 (Conv2D)          (None, 124, 124, 32)      4640      
                                                                 
 conv2d_27 (Conv2D)          (None, 122, 122, 32)      9248      
                                                                 
 batch_normalization_23 (Bat  (None, 122, 122, 32)     128       
 chNormalization)                                                
                                                                 
 average_pooling2d_13 (Avera  (None, 61, 61, 32)       0         
 gePooling2D)                                                    
                                                                 
 dropout_14 (Dropout)        (None, 61, 61, 32)        0         
                                                                 
 conv2d_28 (Conv2D)          (None, 59, 59, 64)        18496     
                                                                 
 conv2d_29 (Conv2D)          (None, 57, 57, 64)        36928     
                                                                 
 batch_normalization_24 (Bat  (None, 57, 57, 64)       256       
 chNormalization)                                                
                                                                 
 average_pooling2d_14 (Avera  (None, 28, 28, 64)       0         
 gePooling2D)                                                    
                                                                 
 dropout_15 (Dropout)        (None, 28, 28, 64)        0         
                                                                 
 conv2d_30 (Conv2D)          (None, 26, 26, 128)       73856     
                                                                 
 conv2d_31 (Conv2D)          (None, 24, 24, 128)       147584    
                                                                 
 batch_normalization_25 (Bat  (None, 24, 24, 128)      512       
 chNormalization)                                                
                                                                 
 dropout_16 (Dropout)        (None, 24, 24, 128)       0         
                                                                 
 conv2d_32 (Conv2D)          (None, 22, 22, 516)       594948    
                                                                 
 conv2d_33 (Conv2D)          (None, 20, 20, 516)       2396820   
                                                                 
 batch_normalization_26 (Bat  (None, 20, 20, 516)      2064      
 chNormalization)                                                
                                                                 
 dropout_17 (Dropout)        (None, 20, 20, 516)       0         
                                                                 
 flatten_4 (Flatten)         (None, 206400)            0         
                                                                 
 dense_8 (Dense)             (None, 516)               106502916 
                                                                 
 batch_normalization_27 (Bat  (None, 516)              2064      
 chNormalization)                                                
                                                                 
 dense_9 (Dense)             (None, 17)                8789      
                                                                 
=================================================================
Total params: 109,802,081
Trainable params: 109,799,537
Non-trainable params: 2,544
_________________________________________________________________

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。

1.设置动态学习率

ExponentialDecay函数：
tf.keras.optimizers.schedules.ExponentialDecay是 TensorFlow 中的一个学习率衰减策略，用于在训练神经网络时动态地降低学习率。学习率衰减是一种常用的技巧，可以帮助优化算法更有效地收敛到全局最小值，从而提高模型的性能。

主要参数：

●initial_learning_rate（初始学习率）：初始学习率大小。
●decay_steps（衰减步数）：学习率衰减的步数。在经过 decay_steps 步后，学习率将按照指数函数衰减。例如，如果 decay_steps 设置为 10，则每10步衰减一次。
●decay_rate（衰减率）：学习率的衰减率。它决定了学习率如何衰减。通常，取值在 0 到 1 之间。
●staircase（阶梯式衰减）：一个布尔值，控制学习率的衰减方式。如果设置为 True，则学习率在每个 decay_steps 步之后直接减小，形成阶梯状下降。如果设置为 False，则学习率将连续衰减。

# 设置初始学习率
initial_learning_rate = 1e-3

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

2.早停与保存最佳模型参数

2.1. ModelCheckpoint 讲解
官网地址：https://www.tensorflow.org/api_docs/python/tf/keras/callbacks/ModelCheckpoint
函数原型：

tf.keras.callbacks.ModelCheckpoint(
    filepath, monitor='val_loss', verbose=0, save_best_only=False,
    save_weights_only=False, mode='auto', save_freq='epoch',
    options=None, **kwargs
)

参数说明：

●filepath：字符串，保存模型的路径。
●monitor：需要监视的值。
●verbose：信息展示模式，0或1(checkpoint的保存信息，类似Epoch 00001: saving model to …)。
●save_best_only：当设置为True时，监测值有改进时才会保存当前的模型（ the latest best model according to the quantity monitored will not be overwritten）。
●mode：‘auto’，‘min’，‘max’之一，在save_best_only=True时决定性能最佳模型的评判准则，例如，当监测值为val_acc时，模式应为max，当监测值为val_loss时，模式应为min。在auto模式下，评价准则由被监测值的名字自动推断。
●save_weights_only：若设置为True，则只保存模型权重，否则将保存整个模型（包括模型结构，配置信息等）。
●period：CheckPoint之间的间隔的epoch数。

2.2. 早停 tf.keras.callbacks.EarlyStopping() 详解
函数原型：

tf.keras.callbacks.EarlyStopping(
    monitor='val_loss', min_delta=0, patience=0, verbose=0,
    mode='auto', baseline=None, restore_best_weights=False
)

官网地址：https://www.tensorflow.org/api_docs/python/tf/keras/callbacks/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。

代码实例：

callback = tf.keras.callbacks.EarlyStopping(monitor='loss', patience=3)
# This callback will stop the training when there is no improvement in
# the loss for three consecutive epochs.
model = tf.keras.models.Sequential([tf.keras.layers.Dense(10)])
model.compile(tf.keras.optimizers.SGD(), loss='mse')
history = model.fit(np.arange(100).reshape(5, 20), np.zeros(5),
                    epochs=10, batch_size=1, callbacks=[callback],
                    verbose=0)
len(history.history['loss'])  # Only 4 epochs are run.

本节正式代码：

from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping

epochs = 100

# 保存最佳模型参数
checkpointer = ModelCheckpoint('T6_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=50, 
                             verbose=1)

3. 模型训练

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

代码输出：

Epoch 1/100
51/51 [==============================] - ETA: 0s - loss: 3.2724 - accuracy: 0.1352
Epoch 1: val_accuracy improved from -inf to 0.04444, saving model to best_model.h5
51/51 [==============================] - 6s 100ms/step - loss: 3.2724 - accuracy: 0.1352 - val_loss: 4.6923 - val_accuracy: 0.0444
Epoch 2/100
51/51 [==============================] - ETA: 0s - loss: 2.3522 - accuracy: 0.2426
Epoch 2: val_accuracy improved from 0.04444 to 0.05556, saving model to best_model.h5
51/51 [==============================] - 5s 95ms/step - loss: 2.3522 - accuracy: 0.2426 - val_loss: 3.9861 - val_accuracy: 0.0556
Epoch 3/100
51/51 [==============================] - ETA: 0s - loss: 1.9500 - accuracy: 0.3617
Epoch 3: val_accuracy did not improve from 0.05556
51/51 [==============================] - 4s 80ms/step - loss: 1.9500 - accuracy: 0.3617 - val_loss: 5.4874 - val_accuracy: 0.0500
Epoch 4/100
51/51 [==============================] - ETA: 0s - loss: 1.6409 - accuracy: 0.4821
Epoch 4: val_accuracy did not improve from 0.05556
51/51 [==============================] - 4s 83ms/step - loss: 1.6409 - accuracy: 0.4821 - val_loss: 4.4026 - val_accuracy: 0.0444
Epoch 5/100
51/51 [==============================] - ETA: 0s - loss: 1.3265 - accuracy: 0.5537
Epoch 5: val_accuracy improved from 0.05556 to 0.10000, saving model to best_model.h5
51/51 [==============================] - 5s 96ms/step - loss: 1.3265 - accuracy: 0.5537 - val_loss: 3.9757 - val_accuracy: 0.1000
Epoch 6/100
51/51 [==============================] - ETA: 0s - loss: 0.9353 - accuracy: 0.7130
Epoch 6: val_accuracy did not improve from 0.10000
51/51 [==============================] - 4s 81ms/step - loss: 0.9353 - accuracy: 0.7130 - val_loss: 3.7666 - val_accuracy: 0.0889
Epoch 7/100
51/51 [==============================] - ETA: 0s - loss: 0.5873 - accuracy: 0.8432
Epoch 7: val_accuracy improved from 0.10000 to 0.13333, saving model to best_model.h5
51/51 [==============================] - 5s 96ms/step - loss: 0.5873 - accuracy: 0.8432 - val_loss: 4.3078 - val_accuracy: 0.1333
Epoch 8/100
51/51 [==============================] - ETA: 0s - loss: 0.3474 - accuracy: 0.9086
Epoch 8: val_accuracy improved from 0.13333 to 0.17222, saving model to best_model.h5
51/51 [==============================] - 5s 95ms/step - loss: 0.3474 - accuracy: 0.9086 - val_loss: 4.2296 - val_accuracy: 0.1722
Epoch 9/100
51/51 [==============================] - ETA: 0s - loss: 0.2292 - accuracy: 0.9488
Epoch 9: val_accuracy improved from 0.17222 to 0.17778, saving model to best_model.h5
51/51 [==============================] - 5s 96ms/step - loss: 0.2292 - accuracy: 0.9488 - val_loss: 4.1251 - val_accuracy: 0.1778
Epoch 10/100
51/51 [==============================] - ETA: 0s - loss: 0.1096 - accuracy: 0.9821
Epoch 10: val_accuracy improved from 0.17778 to 0.25556, saving model to best_model.h5
51/51 [==============================] - 5s 97ms/step - loss: 0.1096 - accuracy: 0.9821 - val_loss: 3.7290 - val_accuracy: 0.2556
Epoch 11/100
51/51 [==============================] - ETA: 0s - loss: 0.0568 - accuracy: 0.9963
Epoch 11: val_accuracy improved from 0.25556 to 0.27778, saving model to best_model.h5
51/51 [==============================] - 5s 96ms/step - loss: 0.0568 - accuracy: 0.9963 - val_loss: 3.7409 - val_accuracy: 0.2778
Epoch 12/100
51/51 [==============================] - ETA: 0s - loss: 0.0311 - accuracy: 0.9963
Epoch 12: val_accuracy improved from 0.27778 to 0.32778, saving model to best_model.h5
51/51 [==============================] - 5s 99ms/step - loss: 0.0311 - accuracy: 0.9963 - val_loss: 3.6441 - val_accuracy: 0.3278
Epoch 13/100
51/51 [==============================] - ETA: 0s - loss: 0.0212 - accuracy: 0.9994
Epoch 13: val_accuracy improved from 0.32778 to 0.34444, saving model to best_model.h5
51/51 [==============================] - 5s 95ms/step - loss: 0.0212 - accuracy: 0.9994 - val_loss: 3.4548 - val_accuracy: 0.3444
Epoch 14/100
50/51 [============================>.] - ETA: 0s - loss: 0.0236 - accuracy: 0.9994
Epoch 14: val_accuracy did not improve from 0.34444
51/51 [==============================] - 4s 81ms/step - loss: 0.0237 - accuracy: 0.9994 - val_loss: 3.5205 - val_accuracy: 0.3111
Epoch 15/100
51/51 [==============================] - ETA: 0s - loss: 0.0203 - accuracy: 0.9988
Epoch 15: val_accuracy improved from 0.34444 to 0.36667, saving model to best_model.h5
51/51 [==============================] - 5s 95ms/step - loss: 0.0203 - accuracy: 0.9988 - val_loss: 3.2594 - val_accuracy: 0.3667
Epoch 16/100
51/51 [==============================] - ETA: 0s - loss: 0.0101 - accuracy: 1.0000
Epoch 16: val_accuracy improved from 0.36667 to 0.37222, saving model to best_model.h5
51/51 [==============================] - 5s 94ms/step - loss: 0.0101 - accuracy: 1.0000 - val_loss: 3.1372 - val_accuracy: 0.3722
Epoch 17/100
51/51 [==============================] - ETA: 0s - loss: 0.0088 - accuracy: 1.0000
Epoch 17: val_accuracy improved from 0.37222 to 0.39444, saving model to best_model.h5
51/51 [==============================] - 5s 95ms/step - loss: 0.0088 - accuracy: 1.0000 - val_loss: 3.2258 - val_accuracy: 0.3944
Epoch 18/100
51/51 [==============================] - ETA: 0s - loss: 0.0072 - accuracy: 1.0000
Epoch 18: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 81ms/step - loss: 0.0072 - accuracy: 1.0000 - val_loss: 3.3096 - val_accuracy: 0.3944
Epoch 19/100
51/51 [==============================] - ETA: 0s - loss: 0.0053 - accuracy: 1.0000
Epoch 19: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 80ms/step - loss: 0.0053 - accuracy: 1.0000 - val_loss: 3.3068 - val_accuracy: 0.3944
Epoch 20/100
51/51 [==============================] - ETA: 0s - loss: 0.0047 - accuracy: 1.0000
Epoch 20: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 83ms/step - loss: 0.0047 - accuracy: 1.0000 - val_loss: 3.3807 - val_accuracy: 0.3833
Epoch 21/100
51/51 [==============================] - ETA: 0s - loss: 0.0042 - accuracy: 1.0000
Epoch 21: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 81ms/step - loss: 0.0042 - accuracy: 1.0000 - val_loss: 3.4177 - val_accuracy: 0.3778
Epoch 22/100
51/51 [==============================] - ETA: 0s - loss: 0.0038 - accuracy: 1.0000
Epoch 22: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 82ms/step - loss: 0.0038 - accuracy: 1.0000 - val_loss: 3.4866 - val_accuracy: 0.3889
Epoch 23/100
51/51 [==============================] - ETA: 0s - loss: 0.0036 - accuracy: 1.0000
Epoch 23: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 80ms/step - loss: 0.0036 - accuracy: 1.0000 - val_loss: 3.4752 - val_accuracy: 0.3944
Epoch 24/100
51/51 [==============================] - ETA: 0s - loss: 0.0031 - accuracy: 1.0000
Epoch 24: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 82ms/step - loss: 0.0031 - accuracy: 1.0000 - val_loss: 3.5228 - val_accuracy: 0.3833
Epoch 25/100
51/51 [==============================] - ETA: 0s - loss: 0.0031 - accuracy: 1.0000
Epoch 25: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 80ms/step - loss: 0.0031 - accuracy: 1.0000 - val_loss: 3.5102 - val_accuracy: 0.3778
Epoch 26/100
51/51 [==============================] - ETA: 0s - loss: 0.0030 - accuracy: 1.0000
Epoch 26: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 82ms/step - loss: 0.0030 - accuracy: 1.0000 - val_loss: 3.5223 - val_accuracy: 0.3944
Epoch 27/100
51/51 [==============================] - ETA: 0s - loss: 0.0026 - accuracy: 1.0000
Epoch 27: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 81ms/step - loss: 0.0026 - accuracy: 1.0000 - val_loss: 3.5806 - val_accuracy: 0.3944
Epoch 28/100
51/51 [==============================] - ETA: 0s - loss: 0.0022 - accuracy: 1.0000
Epoch 28: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 81ms/step - loss: 0.0022 - accuracy: 1.0000 - val_loss: 3.5896 - val_accuracy: 0.3944
Epoch 29/100
51/51 [==============================] - ETA: 0s - loss: 0.0024 - accuracy: 1.0000
Epoch 29: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 81ms/step - loss: 0.0024 - accuracy: 1.0000 - val_loss: 3.6306 - val_accuracy: 0.3833
Epoch 30/100
51/51 [==============================] - ETA: 0s - loss: 0.0023 - accuracy: 1.0000
Epoch 30: val_accuracy did not improve from 0.39444
51/51 [==============================] - 4s 81ms/step - loss: 0.0023 - accuracy: 1.0000 - val_loss: 3.5925 - val_accuracy: 0.3889
Epoch 31/100
51/51 [==============================] - ETA: 0s - loss: 0.0022 - accuracy: 1.0000
Epoch 31: val_accuracy improved from 0.39444 to 0.40000, saving model to best_model.h5
51/51 [==============================] - 5s 99ms/step - loss: 0.0022 - accuracy: 1.0000 - val_loss: 3.6142 - val_accuracy: 0.4000
Epoch 32/100
51/51 [==============================] - ETA: 0s - loss: 0.0021 - accuracy: 1.0000
Epoch 32: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 81ms/step - loss: 0.0021 - accuracy: 1.0000 - val_loss: 3.6612 - val_accuracy: 0.3778
Epoch 33/100
51/51 [==============================] - ETA: 0s - loss: 0.0020 - accuracy: 1.0000
Epoch 33: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 82ms/step - loss: 0.0020 - accuracy: 1.0000 - val_loss: 3.6649 - val_accuracy: 0.3833
Epoch 34/100
51/51 [==============================] - ETA: 0s - loss: 0.0019 - accuracy: 1.0000
Epoch 34: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 82ms/step - loss: 0.0019 - accuracy: 1.0000 - val_loss: 3.6516 - val_accuracy: 0.3833
Epoch 35/100
51/51 [==============================] - ETA: 0s - loss: 0.0020 - accuracy: 1.0000
Epoch 35: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 81ms/step - loss: 0.0020 - accuracy: 1.0000 - val_loss: 3.6446 - val_accuracy: 0.3833
Epoch 36/100
51/51 [==============================] - ETA: 0s - loss: 0.0022 - accuracy: 1.0000
Epoch 36: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 80ms/step - loss: 0.0022 - accuracy: 1.0000 - val_loss: 3.6777 - val_accuracy: 0.3944
Epoch 37/100
51/51 [==============================] - ETA: 0s - loss: 0.0017 - accuracy: 1.0000
Epoch 37: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 81ms/step - loss: 0.0017 - accuracy: 1.0000 - val_loss: 3.6630 - val_accuracy: 0.4000
Epoch 38/100
51/51 [==============================] - ETA: 0s - loss: 0.0018 - accuracy: 1.0000
Epoch 38: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 82ms/step - loss: 0.0018 - accuracy: 1.0000 - val_loss: 3.6627 - val_accuracy: 0.3944
Epoch 39/100
51/51 [==============================] - ETA: 0s - loss: 0.0015 - accuracy: 1.0000
Epoch 39: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 83ms/step - loss: 0.0015 - accuracy: 1.0000 - val_loss: 3.6457 - val_accuracy: 0.4000
Epoch 40/100
51/51 [==============================] - ETA: 0s - loss: 0.0014 - accuracy: 1.0000
Epoch 40: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 84ms/step - loss: 0.0014 - accuracy: 1.0000 - val_loss: 3.6689 - val_accuracy: 0.3944
Epoch 41/100
50/51 [============================>.] - ETA: 0s - loss: 0.0015 - accuracy: 1.0000
Epoch 41: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 83ms/step - loss: 0.0015 - accuracy: 1.0000 - val_loss: 3.7177 - val_accuracy: 0.3889
Epoch 42/100
51/51 [==============================] - ETA: 0s - loss: 0.0014 - accuracy: 1.0000
Epoch 42: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 81ms/step - loss: 0.0014 - accuracy: 1.0000 - val_loss: 3.6984 - val_accuracy: 0.3778
Epoch 43/100
50/51 [============================>.] - ETA: 0s - loss: 0.0013 - accuracy: 1.0000
Epoch 43: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 83ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 3.7101 - val_accuracy: 0.3889
Epoch 44/100
51/51 [==============================] - ETA: 0s - loss: 0.0013 - accuracy: 1.0000
Epoch 44: val_accuracy did not improve from 0.40000
51/51 [==============================] - 4s 82ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 3.7219 - val_accuracy: 0.3833
Epoch 45/100
50/51 [============================>.] - ETA: 0s - loss: 0.0014 - accuracy: 1.0000
Epoch 45: val_accuracy improved from 0.40000 to 0.40556, saving model to best_model.h5
51/51 [==============================] - 5s 98ms/step - loss: 0.0014 - accuracy: 1.0000 - val_loss: 3.7500 - val_accuracy: 0.4056
Epoch 46/100
51/51 [==============================] - ETA: 0s - loss: 0.0013 - accuracy: 1.0000
Epoch 46: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 79ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 3.7347 - val_accuracy: 0.3944
Epoch 47/100
51/51 [==============================] - ETA: 0s - loss: 0.0013 - accuracy: 1.0000
Epoch 47: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 80ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 3.7378 - val_accuracy: 0.3833
Epoch 48/100
51/51 [==============================] - ETA: 0s - loss: 0.0013 - accuracy: 1.0000
Epoch 48: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 84ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 3.7441 - val_accuracy: 0.3722
Epoch 49/100
51/51 [==============================] - ETA: 0s - loss: 0.0012 - accuracy: 1.0000
Epoch 49: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 83ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 3.7446 - val_accuracy: 0.3778
Epoch 50/100
51/51 [==============================] - ETA: 0s - loss: 0.0011 - accuracy: 1.0000
Epoch 50: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 82ms/step - loss: 0.0011 - accuracy: 1.0000 - val_loss: 3.7513 - val_accuracy: 0.3889
Epoch 51/100
51/51 [==============================] - ETA: 0s - loss: 0.0012 - accuracy: 1.0000
Epoch 51: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 83ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 3.7797 - val_accuracy: 0.3833
Epoch 52/100
51/51 [==============================] - ETA: 0s - loss: 0.0011 - accuracy: 1.0000
Epoch 52: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 84ms/step - loss: 0.0011 - accuracy: 1.0000 - val_loss: 3.7682 - val_accuracy: 0.3778
Epoch 53/100
51/51 [==============================] - ETA: 0s - loss: 0.0010 - accuracy: 1.0000
Epoch 53: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 83ms/step - loss: 0.0010 - accuracy: 1.0000 - val_loss: 3.7797 - val_accuracy: 0.3833
Epoch 54/100
51/51 [==============================] - ETA: 0s - loss: 0.0011 - accuracy: 1.0000
Epoch 54: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 81ms/step - loss: 0.0011 - accuracy: 1.0000 - val_loss: 3.7882 - val_accuracy: 0.3833
Epoch 55/100
51/51 [==============================] - ETA: 0s - loss: 9.8179e-04 - accuracy: 1.0000
Epoch 55: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 81ms/step - loss: 9.8179e-04 - accuracy: 1.0000 - val_loss: 3.7911 - val_accuracy: 0.3778
Epoch 56/100
51/51 [==============================] - ETA: 0s - loss: 0.0010 - accuracy: 1.0000
Epoch 56: val_accuracy did not improve from 0.40556
51/51 [==============================] - 4s 81ms/step - loss: 0.0010 - accuracy: 1.0000 - val_loss: 3.7977 - val_accuracy: 0.3833
Epoch 57/100
51/51 [==============================] - ETA: 0s - loss: 0.0010 - accuracy: 1.0000  
Epoch 57: val_accuracy did not improve from 0.40556
51/51 [==============================] - 3s 63ms/step - loss: 0.0010 - accuracy: 1.0000 - val_loss: 3.7906 - val_accuracy: 0.3833
Epoch 58/100
51/51 [==============================] - ETA: 0s - loss: 9.5346e-04 - accuracy: 1.0000
Epoch 58: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 9.5346e-04 - accuracy: 1.0000 - val_loss: 3.7825 - val_accuracy: 0.3889
Epoch 59/100
51/51 [==============================] - ETA: 0s - loss: 9.0027e-04 - accuracy: 1.0000
Epoch 59: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 9.0027e-04 - accuracy: 1.0000 - val_loss: 3.7923 - val_accuracy: 0.3889
Epoch 60/100
51/51 [==============================] - ETA: 0s - loss: 8.7549e-04 - accuracy: 1.0000
Epoch 60: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 8.7549e-04 - accuracy: 1.0000 - val_loss: 3.8056 - val_accuracy: 0.3833
Epoch 61/100
51/51 [==============================] - ETA: 0s - loss: 9.3759e-04 - accuracy: 1.0000
Epoch 61: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 9.3759e-04 - accuracy: 1.0000 - val_loss: 3.8058 - val_accuracy: 0.4000
Epoch 62/100
51/51 [==============================] - ETA: 0s - loss: 9.2041e-04 - accuracy: 1.0000
Epoch 62: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 9.2041e-04 - accuracy: 1.0000 - val_loss: 3.8435 - val_accuracy: 0.3778
Epoch 63/100
51/51 [==============================] - ETA: 0s - loss: 9.1431e-04 - accuracy: 1.0000
Epoch 63: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 9.1431e-04 - accuracy: 1.0000 - val_loss: 3.8381 - val_accuracy: 0.4000
Epoch 64/100
51/51 [==============================] - ETA: 0s - loss: 8.9726e-04 - accuracy: 1.0000
Epoch 64: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 8.9726e-04 - accuracy: 1.0000 - val_loss: 3.8237 - val_accuracy: 0.3889
Epoch 65/100
51/51 [==============================] - ETA: 0s - loss: 9.1290e-04 - accuracy: 1.0000
Epoch 65: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 9.1290e-04 - accuracy: 1.0000 - val_loss: 3.8059 - val_accuracy: 0.3889
Epoch 66/100
51/51 [==============================] - ETA: 0s - loss: 8.4770e-04 - accuracy: 1.0000
Epoch 66: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 8.4770e-04 - accuracy: 1.0000 - val_loss: 3.8043 - val_accuracy: 0.3944
Epoch 67/100
51/51 [==============================] - ETA: 0s - loss: 8.2910e-04 - accuracy: 1.0000
Epoch 67: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 8.2910e-04 - accuracy: 1.0000 - val_loss: 3.8186 - val_accuracy: 0.3944
Epoch 68/100
51/51 [==============================] - ETA: 0s - loss: 8.2363e-04 - accuracy: 1.0000
Epoch 68: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 8.2363e-04 - accuracy: 1.0000 - val_loss: 3.8205 - val_accuracy: 0.3944
Epoch 69/100
51/51 [==============================] - ETA: 0s - loss: 7.5853e-04 - accuracy: 1.0000
Epoch 69: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 7.5853e-04 - accuracy: 1.0000 - val_loss: 3.8187 - val_accuracy: 0.3944
Epoch 70/100
51/51 [==============================] - ETA: 0s - loss: 8.2532e-04 - accuracy: 1.0000
Epoch 70: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 8.2532e-04 - accuracy: 1.0000 - val_loss: 3.8171 - val_accuracy: 0.3944
Epoch 71/100
51/51 [==============================] - ETA: 0s - loss: 8.3162e-04 - accuracy: 1.0000
Epoch 71: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 8.3162e-04 - accuracy: 1.0000 - val_loss: 3.8302 - val_accuracy: 0.3889
Epoch 72/100
51/51 [==============================] - ETA: 0s - loss: 7.3582e-04 - accuracy: 1.0000
Epoch 72: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 7.3582e-04 - accuracy: 1.0000 - val_loss: 3.8317 - val_accuracy: 0.3889
Epoch 73/100
51/51 [==============================] - ETA: 0s - loss: 7.6574e-04 - accuracy: 1.0000
Epoch 73: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 7.6574e-04 - accuracy: 1.0000 - val_loss: 3.8391 - val_accuracy: 0.3944
Epoch 74/100
51/51 [==============================] - ETA: 0s - loss: 7.8223e-04 - accuracy: 1.0000
Epoch 74: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 7.8223e-04 - accuracy: 1.0000 - val_loss: 3.8451 - val_accuracy: 0.3944
Epoch 75/100
51/51 [==============================] - ETA: 0s - loss: 7.2062e-04 - accuracy: 1.0000
Epoch 75: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 7.2062e-04 - accuracy: 1.0000 - val_loss: 3.8480 - val_accuracy: 0.3944
Epoch 76/100
51/51 [==============================] - ETA: 0s - loss: 7.0158e-04 - accuracy: 1.0000
Epoch 76: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 7.0158e-04 - accuracy: 1.0000 - val_loss: 3.8543 - val_accuracy: 0.3944
Epoch 77/100
51/51 [==============================] - ETA: 0s - loss: 6.5803e-04 - accuracy: 1.0000
Epoch 77: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.5803e-04 - accuracy: 1.0000 - val_loss: 3.8559 - val_accuracy: 0.4000
Epoch 78/100
51/51 [==============================] - ETA: 0s - loss: 7.8448e-04 - accuracy: 1.0000
Epoch 78: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 42ms/step - loss: 7.8448e-04 - accuracy: 1.0000 - val_loss: 3.8619 - val_accuracy: 0.4000
Epoch 79/100
51/51 [==============================] - ETA: 0s - loss: 6.8857e-04 - accuracy: 1.0000
Epoch 79: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.8857e-04 - accuracy: 1.0000 - val_loss: 3.8518 - val_accuracy: 0.4056
Epoch 80/100
51/51 [==============================] - ETA: 0s - loss: 7.0843e-04 - accuracy: 1.0000
Epoch 80: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 7.0843e-04 - accuracy: 1.0000 - val_loss: 3.8624 - val_accuracy: 0.3833
Epoch 81/100
51/51 [==============================] - ETA: 0s - loss: 6.8543e-04 - accuracy: 1.0000
Epoch 81: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.8543e-04 - accuracy: 1.0000 - val_loss: 3.8718 - val_accuracy: 0.3833
Epoch 82/100
51/51 [==============================] - ETA: 0s - loss: 6.8276e-04 - accuracy: 1.0000
Epoch 82: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.8276e-04 - accuracy: 1.0000 - val_loss: 3.8830 - val_accuracy: 0.3944
Epoch 83/100
51/51 [==============================] - ETA: 0s - loss: 6.3817e-04 - accuracy: 1.0000
Epoch 83: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.3817e-04 - accuracy: 1.0000 - val_loss: 3.8756 - val_accuracy: 0.3944
Epoch 84/100
51/51 [==============================] - ETA: 0s - loss: 6.7866e-04 - accuracy: 1.0000
Epoch 84: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.7866e-04 - accuracy: 1.0000 - val_loss: 3.8885 - val_accuracy: 0.3944
Epoch 85/100
51/51 [==============================] - ETA: 0s - loss: 6.3098e-04 - accuracy: 1.0000
Epoch 85: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.3098e-04 - accuracy: 1.0000 - val_loss: 3.8852 - val_accuracy: 0.3944
Epoch 86/100
51/51 [==============================] - ETA: 0s - loss: 7.2486e-04 - accuracy: 1.0000
Epoch 86: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 7.2486e-04 - accuracy: 1.0000 - val_loss: 3.9031 - val_accuracy: 0.3889
Epoch 87/100
51/51 [==============================] - ETA: 0s - loss: 5.9727e-04 - accuracy: 1.0000
Epoch 87: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 5.9727e-04 - accuracy: 1.0000 - val_loss: 3.8920 - val_accuracy: 0.3889
Epoch 88/100
51/51 [==============================] - ETA: 0s - loss: 7.0130e-04 - accuracy: 1.0000
Epoch 88: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 7.0130e-04 - accuracy: 1.0000 - val_loss: 3.8863 - val_accuracy: 0.3944
Epoch 89/100
51/51 [==============================] - ETA: 0s - loss: 5.6934e-04 - accuracy: 1.0000
Epoch 89: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 5.6934e-04 - accuracy: 1.0000 - val_loss: 3.8735 - val_accuracy: 0.3944
Epoch 90/100
51/51 [==============================] - ETA: 0s - loss: 6.1416e-04 - accuracy: 1.0000
Epoch 90: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.1416e-04 - accuracy: 1.0000 - val_loss: 3.8717 - val_accuracy: 0.3889
Epoch 91/100
51/51 [==============================] - ETA: 0s - loss: 5.9682e-04 - accuracy: 1.0000
Epoch 91: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 5.9682e-04 - accuracy: 1.0000 - val_loss: 3.8809 - val_accuracy: 0.3889
Epoch 92/100
51/51 [==============================] - ETA: 0s - loss: 6.0320e-04 - accuracy: 1.0000
Epoch 92: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.0320e-04 - accuracy: 1.0000 - val_loss: 3.8958 - val_accuracy: 0.4000
Epoch 93/100
51/51 [==============================] - ETA: 0s - loss: 6.4800e-04 - accuracy: 1.0000
Epoch 93: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.4800e-04 - accuracy: 1.0000 - val_loss: 3.8929 - val_accuracy: 0.3944
Epoch 94/100
51/51 [==============================] - ETA: 0s - loss: 6.3013e-04 - accuracy: 1.0000
Epoch 94: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 6.3013e-04 - accuracy: 1.0000 - val_loss: 3.9022 - val_accuracy: 0.3833
Epoch 95/100
51/51 [==============================] - ETA: 0s - loss: 5.6292e-04 - accuracy: 1.0000
Epoch 95: val_accuracy did not improve from 0.40556
51/51 [==============================] - 2s 41ms/step - loss: 5.6292e-04 - accuracy: 1.0000 - val_loss: 3.9056 - val_accuracy: 0.3889
Epoch 95: early stopping

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()

代码输出：

2. 指定图片进行预测

# 加载效果最好的模型权重
model.load_weights('T6_best_model.h5')

from PIL import Image
import numpy as np

img = Image.open("./T6/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)])

代码输出：

预测结果为： Jennifer Lawrence

3. 随机指定图片进行预测

import random

image_file = list(data_dir.glob('*/*.jpg'))

image_file
image_one=random.choice(image_file)
image_one

代码输出：

PosixPath('T6/data/Hugh Jackman/059_aef6a0a8.jpg')

#要预测的图片
PIL.Image.open(str(image_one))

代码输出：

img = Image.open(image_one)  #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)])

代码输出：

预测结果为： Hugh Jackman

#因为预测的图片和随机给定的图片的标签不一定相同，所以打印预测图片的标签的第一张图片和随机指定的图片进行对比

label_pred=class_names[np.argmax(predictions)]
label_pred

image_pred = list(data_dir.glob(label_pred+'/'+'*.jpg'))
image_pred
PIL.Image.open(str(image_pred[0]))

代码输出：

六、总结

从训练结果可知，accuracy的准确率很高，可达100%，而val_accuracy最高是40%，过拟合严重。
为了提高val_accuracy，尝试了用自己搭建的VGG-16网络框架和调用官方的VGG-16网络框架来训练该模型。调用官方的VGG-16网络框架来训练该模型时val_accuracy是最高的，基本在55%，偶尔一次训练在60%。
还是要改进模型网络来提高val_accuracy。
至于怎么搭建的VGG-16网络框架，怎么调用官方的VGG-16网络框架，可参考同系列的后面的几遍文章，后面的几遍文章均有描述VGG-16网络框架。

七、别人的总结，可以借鉴学习：

1.直接调用VGG16
效果其实很差，过拟合严重，于是通过以下手段调优：

1）添加BN层：有效。
2）Drop层：卷积层无效，甚至有反效果，全连接层有效。
3）L2正则化：有效，但参数不能设置很高。
4）减少全连接层神经元数量。
5) 调整图片尺寸，224x224------256x256

上述操作结束，已经可以达到0.5左右，最高大概0.56。

2. 加载预训练参数

1.自己搭建的VGG16有BN层，无法直接加载参数，只能含泪注释掉。
2.加载前13层，并冻结参数，减少顶层神经元，并加入BN层和Drop层，可以有效提升精度，大概20多个点，最终精度可以稳步达到0.7以上，跑了几次，最高能达到0.76+。