基于tensorflow的好莱坞明星识别

最新推荐文章于 2024-07-18 18:43:19 发布

Tooii

最新推荐文章于 2024-07-18 18:43:19 发布

阅读量582

点赞数 30

分类专栏： tensorflow pytorch 文章标签： tensorflow 人工智能 python 分类

本文链接：https://blog.csdn.net/weixin_53657683/article/details/138561433

版权

pytorch 同时被 2 个专栏收录

11 篇文章 0 订阅

订阅专栏

tensorflow

8 篇文章 0 订阅

订阅专栏

🍨 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

一、前期工作

1. 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

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

2. 导入数据

data_dir = "F:/host/Data/好莱坞明星人脸识别识别数据/"

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)

图片总数为： 1800

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

在这里插入图片描述

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）。

"""
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)

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)

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)

['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")

在这里插入图片描述

3. 再次检查数据

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

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

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

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)

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。

"""
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

Model: "sequential_1"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 rescaling_1 (Rescaling)     (None, 224, 224, 3)       0         
                                                                 
 conv2d_4 (Conv2D)           (None, 222, 222, 16)      448       
                                                                 
 average_pooling2d_3 (Averag  (None, 111, 111, 16)     0         
 ePooling2D)                                                     
                                                                 
 conv2d_5 (Conv2D)           (None, 109, 109, 32)      4640      
                                                                 
 average_pooling2d_4 (Averag  (None, 54, 54, 32)       0         
 ePooling2D)                                                     
                                                                 
 dropout_3 (Dropout)         (None, 54, 54, 32)        0         
                                                                 
 conv2d_6 (Conv2D)           (None, 52, 52, 64)        18496     
                                                                 
 average_pooling2d_5 (Averag  (None, 26, 26, 64)       0         
 ePooling2D)                                                     
                                                                 
 dropout_4 (Dropout)         (None, 26, 26, 64)        0         
                                                                 
 conv2d_7 (Conv2D)           (None, 24, 24, 128)       73856     
                                                                 
 dropout_5 (Dropout)         (None, 24, 24, 128)       0         
                                                                 
 flatten_1 (Flatten)         (None, 73728)             0         
                                                                 
 dense_2 (Dense)             (None, 128)               9437312   
                                                                 
 dense_3 (Dense)             (None, 17)                2193      
                                                                 
=================================================================
Total params: 9,536,945
Trainable params: 9,536,945
Non-trainable params: 0
_________________________________________________________________

四、训练模型

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'])

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（对数损失函数）

与 softmax 相对应的损失函数，针对于多分类问题，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=optimizer,
              loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
              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，针对于多分类问题，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=optimizer,
                  loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
                  metrics=['accuracy'] 
    )

函数原型

    tf.keras.losses.SparseCategoricalCrossentropy(
        from_logits=False, 
        reduction=losses.Reduction.AUTO, 
        name='sparse_categorical_crossentropy')

参数说明

from_logits：是否将预测值转换为概率值，即是否使用softmax函数。

如果为True，则将预测值转换为概率值，即使用softmax函数。

如果为False，则不进行转换，直接使用预测值作为损失函数的输入。

默认值为False。
reduction：损失函数的计算方式。

可选值为：
- losses.Reduction.SUM：将所有样本的损失值相加，然后除以样本数量。
- losses.Reduction.NONE：不进行任何归一化操作，直接返回每个样本的损失值。
- losses.Reduction.AUTO：根据输入的形状自动选择计算方式。
默认值为losses.Reduction.AUTO。
name：损失函数的名称。

默认值为sparse_categorical_crossentropy。

2.早停与保存最佳模型参数

关于ModelCheckpoint的详细介绍可参考文章：https://blog.csdn.net/qq_38251616/article/details/122318957

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。

关于EarlyStopping()的详细介绍可参考文章：https://blog.csdn.net/qq_38251616/article/details/122319538

from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping

epochs = 100

# 保存最佳模型参数
checkpointer = ModelCheckpoint('./model/好莱坞明星识别_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])

Epoch 1/100
50/51 [============================>.] - ETA: 0s - loss: 2.8168 - accuracy: 0.1064
Epoch 1: val_accuracy improved from -inf to 0.14444, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 4s 40ms/step - loss: 2.8174 - accuracy: 0.1056 - val_loss: 2.8029 - val_accuracy: 0.1444
Epoch 2/100
50/51 [============================>.] - ETA: 0s - loss: 2.7704 - accuracy: 0.1159
Epoch 2: val_accuracy did not improve from 0.14444
51/51 [==============================] - 2s 31ms/step - loss: 2.7691 - accuracy: 0.1154 - val_loss: 2.7592 - val_accuracy: 0.1444
Epoch 3/100
50/51 [============================>.] - ETA: 0s - loss: 2.7197 - accuracy: 0.1379
Epoch 3: val_accuracy improved from 0.14444 to 0.21667, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 33ms/step - loss: 2.7186 - accuracy: 0.1377 - val_loss: 2.6696 - val_accuracy: 0.2167
Epoch 4/100
50/51 [============================>.] - ETA: 0s - loss: 2.5964 - accuracy: 0.1606
Epoch 4: val_accuracy did not improve from 0.21667
51/51 [==============================] - 2s 31ms/step - loss: 2.5905 - accuracy: 0.1648 - val_loss: 2.5789 - val_accuracy: 0.2056
Epoch 5/100
51/51 [==============================] - ETA: 0s - loss: 2.4681 - accuracy: 0.1975
Epoch 5: val_accuracy did not improve from 0.21667
51/51 [==============================] - 2s 31ms/step - loss: 2.4681 - accuracy: 0.1975 - val_loss: 2.5502 - val_accuracy: 0.1500
Epoch 6/100
50/51 [============================>.] - ETA: 0s - loss: 2.3424 - accuracy: 0.2361
Epoch 6: val_accuracy improved from 0.21667 to 0.24444, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 32ms/step - loss: 2.3382 - accuracy: 0.2364 - val_loss: 2.4447 - val_accuracy: 0.2444
Epoch 7/100
50/51 [============================>.] - ETA: 0s - loss: 2.2089 - accuracy: 0.2727
Epoch 7: val_accuracy did not improve from 0.24444
51/51 [==============================] - 2s 31ms/step - loss: 2.2096 - accuracy: 0.2710 - val_loss: 2.4229 - val_accuracy: 0.1944
Epoch 8/100
50/51 [============================>.] - ETA: 0s - loss: 2.1083 - accuracy: 0.3312
Epoch 8: val_accuracy did not improve from 0.24444
51/51 [==============================] - 2s 31ms/step - loss: 2.1141 - accuracy: 0.3296 - val_loss: 2.3777 - val_accuracy: 0.2167
Epoch 9/100
51/51 [==============================] - ETA: 0s - loss: 2.0169 - accuracy: 0.3302
Epoch 9: val_accuracy did not improve from 0.24444
51/51 [==============================] - 2s 31ms/step - loss: 2.0169 - accuracy: 0.3302 - val_loss: 2.4826 - val_accuracy: 0.2222
Epoch 10/100
50/51 [============================>.] - ETA: 0s - loss: 1.9355 - accuracy: 0.3703
Epoch 10: val_accuracy did not improve from 0.24444
51/51 [==============================] - 2s 32ms/step - loss: 1.9361 - accuracy: 0.3691 - val_loss: 2.3449 - val_accuracy: 0.2333
Epoch 11/100
50/51 [============================>.] - ETA: 0s - loss: 1.8232 - accuracy: 0.4144
Epoch 11: val_accuracy did not improve from 0.24444
51/51 [==============================] - 2s 31ms/step - loss: 1.8288 - accuracy: 0.4130 - val_loss: 2.3971 - val_accuracy: 0.2222
Epoch 12/100
50/51 [============================>.] - ETA: 0s - loss: 1.7745 - accuracy: 0.4263
Epoch 12: val_accuracy improved from 0.24444 to 0.28333, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 33ms/step - loss: 1.7787 - accuracy: 0.4265 - val_loss: 2.4528 - val_accuracy: 0.2833
Epoch 13/100
50/51 [============================>.] - ETA: 0s - loss: 1.6931 - accuracy: 0.4540
Epoch 13: val_accuracy improved from 0.28333 to 0.31667, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 33ms/step - loss: 1.6865 - accuracy: 0.4574 - val_loss: 2.4144 - val_accuracy: 0.3167
Epoch 14/100
51/51 [==============================] - ETA: 0s - loss: 1.5903 - accuracy: 0.5000
Epoch 14: val_accuracy did not improve from 0.31667
51/51 [==============================] - 2s 31ms/step - loss: 1.5903 - accuracy: 0.5000 - val_loss: 2.4333 - val_accuracy: 0.3000
Epoch 15/100
50/51 [============================>.] - ETA: 0s - loss: 1.5582 - accuracy: 0.5113
Epoch 15: val_accuracy did not improve from 0.31667
51/51 [==============================] - 2s 31ms/step - loss: 1.5563 - accuracy: 0.5117 - val_loss: 2.3228 - val_accuracy: 0.3000
Epoch 16/100
50/51 [============================>.] - ETA: 0s - loss: 1.4643 - accuracy: 0.5378
Epoch 16: val_accuracy improved from 0.31667 to 0.32778, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 33ms/step - loss: 1.4646 - accuracy: 0.5370 - val_loss: 2.3622 - val_accuracy: 0.3278
Epoch 17/100
50/51 [============================>.] - ETA: 0s - loss: 1.4028 - accuracy: 0.5542
Epoch 17: val_accuracy improved from 0.32778 to 0.33889, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 33ms/step - loss: 1.3961 - accuracy: 0.5586 - val_loss: 2.3537 - val_accuracy: 0.3389
Epoch 18/100
50/51 [============================>.] - ETA: 0s - loss: 1.3079 - accuracy: 0.5970
Epoch 18: val_accuracy improved from 0.33889 to 0.35556, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 33ms/step - loss: 1.3109 - accuracy: 0.5981 - val_loss: 2.3690 - val_accuracy: 0.3556
Epoch 19/100
50/51 [============================>.] - ETA: 0s - loss: 1.2587 - accuracy: 0.6026
Epoch 19: val_accuracy did not improve from 0.35556
51/51 [==============================] - 2s 31ms/step - loss: 1.2683 - accuracy: 0.5994 - val_loss: 2.4178 - val_accuracy: 0.3222
Epoch 20/100
51/51 [==============================] - ETA: 0s - loss: 1.1746 - accuracy: 0.6426
Epoch 20: val_accuracy did not improve from 0.35556
51/51 [==============================] - 2s 32ms/step - loss: 1.1746 - accuracy: 0.6426 - val_loss: 2.4300 - val_accuracy: 0.3056
Epoch 21/100
51/51 [==============================] - ETA: 0s - loss: 1.1117 - accuracy: 0.6488
Epoch 21: val_accuracy did not improve from 0.35556
51/51 [==============================] - 2s 32ms/step - loss: 1.1117 - accuracy: 0.6488 - val_loss: 2.4838 - val_accuracy: 0.3556
Epoch 22/100
51/51 [==============================] - ETA: 0s - loss: 1.0289 - accuracy: 0.6877
Epoch 22: val_accuracy improved from 0.35556 to 0.37222, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 34ms/step - loss: 1.0289 - accuracy: 0.6877 - val_loss: 2.5050 - val_accuracy: 0.3722
Epoch 23/100
51/51 [==============================] - ETA: 0s - loss: 1.0267 - accuracy: 0.6778
Epoch 23: val_accuracy did not improve from 0.37222
51/51 [==============================] - 2s 32ms/step - loss: 1.0267 - accuracy: 0.6778 - val_loss: 2.4621 - val_accuracy: 0.3333
Epoch 24/100
51/51 [==============================] - ETA: 0s - loss: 0.9193 - accuracy: 0.7074
Epoch 24: val_accuracy did not improve from 0.37222
51/51 [==============================] - 2s 33ms/step - loss: 0.9193 - accuracy: 0.7074 - val_loss: 2.4800 - val_accuracy: 0.3667
Epoch 25/100
51/51 [==============================] - ETA: 0s - loss: 0.8588 - accuracy: 0.7315
Epoch 25: val_accuracy did not improve from 0.37222
51/51 [==============================] - 2s 33ms/step - loss: 0.8588 - accuracy: 0.7315 - val_loss: 2.4512 - val_accuracy: 0.3722
Epoch 26/100
51/51 [==============================] - ETA: 0s - loss: 0.8278 - accuracy: 0.7272
Epoch 26: val_accuracy did not improve from 0.37222
51/51 [==============================] - 2s 33ms/step - loss: 0.8278 - accuracy: 0.7272 - val_loss: 2.6059 - val_accuracy: 0.3500
Epoch 27/100
51/51 [==============================] - ETA: 0s - loss: 0.7679 - accuracy: 0.7630
Epoch 27: val_accuracy did not improve from 0.37222
51/51 [==============================] - 2s 33ms/step - loss: 0.7679 - accuracy: 0.7630 - val_loss: 2.5223 - val_accuracy: 0.3722
Epoch 28/100
51/51 [==============================] - ETA: 0s - loss: 0.7020 - accuracy: 0.7722
Epoch 28: val_accuracy did not improve from 0.37222
51/51 [==============================] - 2s 34ms/step - loss: 0.7020 - accuracy: 0.7722 - val_loss: 2.6698 - val_accuracy: 0.3667
Epoch 29/100
51/51 [==============================] - ETA: 0s - loss: 0.6614 - accuracy: 0.7926
Epoch 29: val_accuracy improved from 0.37222 to 0.40000, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 35ms/step - loss: 0.6614 - accuracy: 0.7926 - val_loss: 2.5832 - val_accuracy: 0.4000
Epoch 30/100
51/51 [==============================] - ETA: 0s - loss: 0.6528 - accuracy: 0.7988
Epoch 30: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 33ms/step - loss: 0.6528 - accuracy: 0.7988 - val_loss: 2.6500 - val_accuracy: 0.3722
Epoch 31/100
51/51 [==============================] - ETA: 0s - loss: 0.5936 - accuracy: 0.8167
Epoch 31: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 33ms/step - loss: 0.5936 - accuracy: 0.8167 - val_loss: 2.7787 - val_accuracy: 0.3889
Epoch 32/100
51/51 [==============================] - ETA: 0s - loss: 0.5690 - accuracy: 0.8235
Epoch 32: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 34ms/step - loss: 0.5690 - accuracy: 0.8235 - val_loss: 2.7991 - val_accuracy: 0.3611
Epoch 33/100
51/51 [==============================] - ETA: 0s - loss: 0.5228 - accuracy: 0.8432
Epoch 33: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 34ms/step - loss: 0.5228 - accuracy: 0.8432 - val_loss: 2.8537 - val_accuracy: 0.3722
Epoch 34/100
51/51 [==============================] - ETA: 0s - loss: 0.5079 - accuracy: 0.8475
Epoch 34: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 34ms/step - loss: 0.5079 - accuracy: 0.8475 - val_loss: 2.7718 - val_accuracy: 0.4000
Epoch 35/100
51/51 [==============================] - ETA: 0s - loss: 0.4884 - accuracy: 0.8438
Epoch 35: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 34ms/step - loss: 0.4884 - accuracy: 0.8438 - val_loss: 2.8398 - val_accuracy: 0.3611
Epoch 36/100
51/51 [==============================] - ETA: 0s - loss: 0.4437 - accuracy: 0.8753
Epoch 36: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 34ms/step - loss: 0.4437 - accuracy: 0.8753 - val_loss: 2.8568 - val_accuracy: 0.3944
Epoch 37/100
51/51 [==============================] - ETA: 0s - loss: 0.4274 - accuracy: 0.8623
Epoch 37: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 34ms/step - loss: 0.4274 - accuracy: 0.8623 - val_loss: 2.8914 - val_accuracy: 0.3889
Epoch 38/100
51/51 [==============================] - ETA: 0s - loss: 0.3903 - accuracy: 0.8765
Epoch 38: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 34ms/step - loss: 0.3903 - accuracy: 0.8765 - val_loss: 2.9627 - val_accuracy: 0.3667
Epoch 39/100
51/51 [==============================] - ETA: 0s - loss: 0.3692 - accuracy: 0.8833
Epoch 39: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 34ms/step - loss: 0.3692 - accuracy: 0.8833 - val_loss: 3.0202 - val_accuracy: 0.3889
Epoch 40/100
51/51 [==============================] - ETA: 0s - loss: 0.3492 - accuracy: 0.8981
Epoch 40: val_accuracy did not improve from 0.40000
51/51 [==============================] - 2s 34ms/step - loss: 0.3492 - accuracy: 0.8981 - val_loss: 3.1077 - val_accuracy: 0.3889
Epoch 41/100
51/51 [==============================] - ETA: 0s - loss: 0.3307 - accuracy: 0.9037
Epoch 41: val_accuracy improved from 0.40000 to 0.41111, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 36ms/step - loss: 0.3307 - accuracy: 0.9037 - val_loss: 3.1079 - val_accuracy: 0.4111
Epoch 42/100
51/51 [==============================] - ETA: 0s - loss: 0.3219 - accuracy: 0.8994
Epoch 42: val_accuracy did not improve from 0.41111
51/51 [==============================] - 2s 36ms/step - loss: 0.3219 - accuracy: 0.8994 - val_loss: 3.0905 - val_accuracy: 0.3944
Epoch 43/100
51/51 [==============================] - ETA: 0s - loss: 0.2907 - accuracy: 0.9179
Epoch 43: val_accuracy did not improve from 0.41111
51/51 [==============================] - 2s 35ms/step - loss: 0.2907 - accuracy: 0.9179 - val_loss: 3.0915 - val_accuracy: 0.4000
Epoch 44/100
51/51 [==============================] - ETA: 0s - loss: 0.3048 - accuracy: 0.9154
Epoch 44: val_accuracy did not improve from 0.41111
51/51 [==============================] - 2s 35ms/step - loss: 0.3048 - accuracy: 0.9154 - val_loss: 3.1775 - val_accuracy: 0.3944
Epoch 45/100
51/51 [==============================] - ETA: 0s - loss: 0.2953 - accuracy: 0.9117
Epoch 45: val_accuracy did not improve from 0.41111
51/51 [==============================] - 2s 36ms/step - loss: 0.2953 - accuracy: 0.9117 - val_loss: 3.1419 - val_accuracy: 0.3889
Epoch 46/100
51/51 [==============================] - ETA: 0s - loss: 0.2644 - accuracy: 0.9340
Epoch 46: val_accuracy did not improve from 0.41111
51/51 [==============================] - 2s 35ms/step - loss: 0.2644 - accuracy: 0.9340 - val_loss: 3.2602 - val_accuracy: 0.3944
Epoch 47/100
51/51 [==============================] - ETA: 0s - loss: 0.2554 - accuracy: 0.9296
Epoch 47: val_accuracy did not improve from 0.41111
51/51 [==============================] - 2s 35ms/step - loss: 0.2554 - accuracy: 0.9296 - val_loss: 3.3103 - val_accuracy: 0.3833
Epoch 48/100
51/51 [==============================] - ETA: 0s - loss: 0.2667 - accuracy: 0.9167
Epoch 48: val_accuracy improved from 0.41111 to 0.42222, saving model to ./model\好莱坞明星识别_best_model.h5
51/51 [==============================] - 2s 37ms/step - loss: 0.2667 - accuracy: 0.9167 - val_loss: 3.1441 - val_accuracy: 0.4222
Epoch 49/100
51/51 [==============================] - ETA: 0s - loss: 0.2386 - accuracy: 0.9302
Epoch 49: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 35ms/step - loss: 0.2386 - accuracy: 0.9302 - val_loss: 3.2786 - val_accuracy: 0.3833
Epoch 50/100
51/51 [==============================] - ETA: 0s - loss: 0.2260 - accuracy: 0.9358
Epoch 50: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 35ms/step - loss: 0.2260 - accuracy: 0.9358 - val_loss: 3.2839 - val_accuracy: 0.3889
Epoch 51/100
51/51 [==============================] - ETA: 0s - loss: 0.2175 - accuracy: 0.9383
Epoch 51: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 35ms/step - loss: 0.2175 - accuracy: 0.9383 - val_loss: 3.3833 - val_accuracy: 0.3833
Epoch 52/100
51/51 [==============================] - ETA: 0s - loss: 0.2225 - accuracy: 0.9315
Epoch 52: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 36ms/step - loss: 0.2225 - accuracy: 0.9315 - val_loss: 3.3515 - val_accuracy: 0.3889
Epoch 53/100
51/51 [==============================] - ETA: 0s - loss: 0.1964 - accuracy: 0.9463
Epoch 53: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 36ms/step - loss: 0.1964 - accuracy: 0.9463 - val_loss: 3.3062 - val_accuracy: 0.4056
Epoch 54/100
51/51 [==============================] - ETA: 0s - loss: 0.2013 - accuracy: 0.9438
Epoch 54: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 37ms/step - loss: 0.2013 - accuracy: 0.9438 - val_loss: 3.3462 - val_accuracy: 0.3889
Epoch 55/100
51/51 [==============================] - ETA: 0s - loss: 0.1928 - accuracy: 0.9488
Epoch 55: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 35ms/step - loss: 0.1928 - accuracy: 0.9488 - val_loss: 3.4539 - val_accuracy: 0.3889
Epoch 56/100
51/51 [==============================] - ETA: 0s - loss: 0.1976 - accuracy: 0.9475
Epoch 56: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 37ms/step - loss: 0.1976 - accuracy: 0.9475 - val_loss: 3.4664 - val_accuracy: 0.3889
Epoch 57/100
51/51 [==============================] - ETA: 0s - loss: 0.1866 - accuracy: 0.9475
Epoch 57: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 37ms/step - loss: 0.1866 - accuracy: 0.9475 - val_loss: 3.4143 - val_accuracy: 0.3833
Epoch 58/100
51/51 [==============================] - ETA: 0s - loss: 0.2077 - accuracy: 0.9407
Epoch 58: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 35ms/step - loss: 0.2077 - accuracy: 0.9407 - val_loss: 3.5146 - val_accuracy: 0.4111
Epoch 59/100
51/51 [==============================] - ETA: 0s - loss: 0.1606 - accuracy: 0.9568
Epoch 59: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 35ms/step - loss: 0.1606 - accuracy: 0.9568 - val_loss: 3.4810 - val_accuracy: 0.3889
Epoch 60/100
51/51 [==============================] - ETA: 0s - loss: 0.1473 - accuracy: 0.9593
Epoch 60: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 36ms/step - loss: 0.1473 - accuracy: 0.9593 - val_loss: 3.4740 - val_accuracy: 0.4167
Epoch 61/100
51/51 [==============================] - ETA: 0s - loss: 0.1676 - accuracy: 0.9562
Epoch 61: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 38ms/step - loss: 0.1676 - accuracy: 0.9562 - val_loss: 3.4758 - val_accuracy: 0.4000
Epoch 62/100
51/51 [==============================] - ETA: 0s - loss: 0.1636 - accuracy: 0.9500
Epoch 62: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 36ms/step - loss: 0.1636 - accuracy: 0.9500 - val_loss: 3.5540 - val_accuracy: 0.3833
Epoch 63/100
51/51 [==============================] - ETA: 0s - loss: 0.1680 - accuracy: 0.9549
Epoch 63: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 36ms/step - loss: 0.1680 - accuracy: 0.9549 - val_loss: 3.5183 - val_accuracy: 0.3944
Epoch 64/100
51/51 [==============================] - ETA: 0s - loss: 0.1541 - accuracy: 0.9568
Epoch 64: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 36ms/step - loss: 0.1541 - accuracy: 0.9568 - val_loss: 3.5818 - val_accuracy: 0.4056
Epoch 65/100
51/51 [==============================] - ETA: 0s - loss: 0.1491 - accuracy: 0.9636
Epoch 65: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 36ms/step - loss: 0.1491 - accuracy: 0.9636 - val_loss: 3.5810 - val_accuracy: 0.3944
Epoch 66/100
51/51 [==============================] - ETA: 0s - loss: 0.1423 - accuracy: 0.9630
Epoch 66: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 36ms/step - loss: 0.1423 - accuracy: 0.9630 - val_loss: 3.4993 - val_accuracy: 0.4056
Epoch 67/100
51/51 [==============================] - ETA: 0s - loss: 0.1450 - accuracy: 0.9605
Epoch 67: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 36ms/step - loss: 0.1450 - accuracy: 0.9605 - val_loss: 3.4847 - val_accuracy: 0.4167
Epoch 68/100
51/51 [==============================] - ETA: 0s - loss: 0.1476 - accuracy: 0.9531
Epoch 68: val_accuracy did not improve from 0.42222
51/51 [==============================] - 2s 37ms/step - loss: 0.1476 - accuracy: 0.9531 - val_loss: 3.5448 - val_accuracy: 0.4111
Epoch 68: 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()

在这里插入图片描述

2. 指定图片进行预测

# 加载效果最好的模型权重
model.load_weights('./model/好莱坞明星识别_best_model.h5')

from PIL import Image
import numpy as np
import tensorflow as tf

# 加载图像并调整大小
img = Image.open("F:/host/Data/好莱坞明星人脸识别识别数据/Tom Cruise/004_dc64d954.jpg")
img = img.resize((img_width, img_height))  # 调整图像大小以匹配模型输入尺寸
img_array = np.array(img) / 255.0  # 将图像转换为 NumPy 数组并归一化

# 将 NumPy 数组转换为 TensorFlow 张量
img_tensor = tf.convert_to_tensor(img_array, dtype=tf.float32)
img_tensor = tf.expand_dims(img_tensor, axis=0)  # 在第一个维度上添加批处理维度

# 进行预测
predictions = model.predict(img_tensor)
predicted_class = np.argmax(predictions)

print("预测结果为：", class_names[predicted_class])

1/1 [==============================] - 0s 22ms/step
预测结果为： Tom Cruise

加载效果最好的模型权重
model.load_weights(‘./model/好莱坞明星识别_best_model.h5’)



```python
from PIL import Image
import numpy as np
import tensorflow as tf

# 加载图像并调整大小
img = Image.open("F:/host/Data/好莱坞明星人脸识别识别数据/Tom Cruise/004_dc64d954.jpg")
img = img.resize((img_width, img_height))  # 调整图像大小以匹配模型输入尺寸
img_array = np.array(img) / 255.0  # 将图像转换为 NumPy 数组并归一化

# 将 NumPy 数组转换为 TensorFlow 张量
img_tensor = tf.convert_to_tensor(img_array, dtype=tf.float32)
img_tensor = tf.expand_dims(img_tensor, axis=0)  # 在第一个维度上添加批处理维度

# 进行预测
predictions = model.predict(img_tensor)
predicted_class = np.argmax(predictions)

print("预测结果为：", class_names[predicted_class])

1/1 [==============================] - 0s 22ms/step
预测结果为： Tom Cruise

个人收获

通过本次实践，我对深度学习中的环境配置、数据预处理、模型构建、训练优化以及评估预测等关键步骤有了更深入的理解和实践经验。我学会了如何高效利用GPU资源，掌握了数据集的加载和预处理技巧，以及如何通过调整卷积神经网络的结构和参数来提升模型性能。此外，我也认识到了Dropout和学习率调度在防止过拟合和加快收敛中的重要性，并通过实际操作加深了对早停法等回调函数的运用

Tooii

关注

30
点赞
踩
25

收藏

觉得还不错? 一键收藏
0
评论
基于tensorflow的好莱坞明星识别

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。我学会了如何高效利用GPU资源，掌握了数据集的加载和预处理技巧，以及如何通过调整卷积神经网络的结构和参数来提升模型性能。关于EarlyStopping()的详细介绍可参考文章：https://blog.csdn.net/qq_38251616/article/details/122319538。：损失函数的计算方式。
复制链接

扫一扫