第T5周：运动鞋品牌识别

• 🍨 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

我的环境

• 操作系统：CentOS7
• 显卡：RTX3090 两张
• 显卡驱动：550.78
• CUDA版本: 12.4
• 语言环境：Python3.9.19
• 编译器：Jupyter Lab
• 深度学习环境：
  • TensorFlow-2.17.0 (GPU版本)

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

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'), PhysicalDevice(name='/physical_device:GPU:1', device_type='GPU')]

2. 导入数据

data_dir = "./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： 578

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

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

• tf.keras.preprocessing.image_dataset_from_directory()：是 TensorFlow 的 Keras 模块中的一个函数，用于从目录中创建一个图像数据集（dataset）。这个函数可以以更方便的方式加载图像数据，用于训练和评估神经网络模型。

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 = 224
img_width = 224

如果准备尝试 categorical_crossentropy损失函数，下面的代码遇到变动哈，变动细节将在下一周博客内公布。

"""
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/train/",
    seed=123,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 502 files belonging to 2 classes.


2024-10-02 19:43:02.432670: I tensorflow/core/common_runtime/gpu/gpu_device.cc:2021] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 21983 MB memory:  -> device: 0, name: NVIDIA GeForce RTX 3090, pci bus id: 0000:3b:00.0, compute capability: 8.6

"""
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/test/",
    seed=123,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 76 files belonging to 2 classes.

class_names = train_ds.class_names
print(class_names)

['adidas', 'nike']

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[labels[i]])
        
        plt.axis("off")

2024-10-02 19:44:06.137769: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence

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

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

• 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

3. 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)

• shuffle() ：打乱数据，关于此函数的详细介绍可以参考：zhuanlan.zhihu.com/p/42417456
• prefetch() ：预取数据，加速运行
• cache() : 将数据集缓存到内存当中，加速运行

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。

• 网络结构图：

"""
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.AveragePooling2D((2, 2)),               # 池化层1，2*2采样
    layers.MaxPooling2D((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.MaxPooling2D((2, 2)),               # 池化层2，2*2采样
    #layers.Dropout(0.3),  
    layers.Conv2D(64, (3, 3), activation='relu'),  # 卷积层3，卷积核3*3
    layers.Dropout(0.3),  
    
    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_7"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ rescaling_7 (Rescaling)         │ (None, 224, 224, 3)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_21 (Conv2D)              │ (None, 222, 222, 16)   │           448 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_4 (MaxPooling2D)  │ (None, 111, 111, 16)   │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_22 (Conv2D)              │ (None, 109, 109, 32)   │         4,640 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_5 (MaxPooling2D)  │ (None, 54, 54, 32)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_23 (Conv2D)              │ (None, 52, 52, 64)     │        18,496 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout_11 (Dropout)            │ (None, 52, 52, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ flatten_7 (Flatten)             │ (None, 173056)         │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_14 (Dense)                │ (None, 128)            │    22,151,296 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_15 (Dense)                │ (None, 2)              │           258 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 22,175,138 (84.59 MB)

 Trainable params: 22,175,138 (84.59 MB)

 Non-trainable params: 0 (0.00 B)

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。

注：这里设置的动态学习率为：指数衰减型（ExponentialDecay）。在每一个epoch开始前，学习率（learning_rate）都将会重置为初始学习率（initial_learning_rate），然后再重新开始衰减。计算公式如下：

learning_rate = initial_learning_rate * decay_rate ^ (step / decay_steps)

学习率大与学习率小的优缺点分析:

学习率大

• 优点： 1、加快学习速率。 2、有助于跳出局部最优值。
• 缺点： 1、导致模型训练不收敛。 2、单单使用大学习率容易导致模型不精确。

学习率小

• 优点： 1、有助于模型收敛、模型细化。 2、提高模型精度。
• 缺点： 1、很难跳出局部最优值。 2、收敛缓慢。

# 设置初始学习率
initial_learning_rate = 0.001

lr_schedule = tf.keras.optimizers.schedules.ExponentialDecay(
        initial_learning_rate, 
        decay_steps=10,      # 敲黑板！！！这里是指 steps，不是指epochs
        decay_rate=0.92,     # 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.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

2. 保存最佳模型

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)

3. 模型训练

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

Epoch 1/100
13/16 ━━━━━━━━━━━━━━━━━━━━ 2s 795ms/step - accuracy: 0.5011 - loss: 2.8015
Epoch 1: val_accuracy improved from -inf to 0.53947, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 20s 891ms/step - accuracy: 0.5004 - loss: 2.5806 - val_accuracy: 0.5395 - val_loss: 0.6970
Epoch 2/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 11ms/step - accuracy: 0.5947 - loss: 0.6634
Epoch 2: val_accuracy improved from 0.53947 to 0.63158, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 2s 116ms/step - accuracy: 0.5938 - loss: 0.6637 - val_accuracy: 0.6316 - val_loss: 0.6295
Epoch 3/100
14/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 0.6631 - loss: 0.6086
Epoch 3: val_accuracy improved from 0.63158 to 0.68421, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 2s 108ms/step - accuracy: 0.6660 - loss: 0.6072 - val_accuracy: 0.6842 - val_loss: 0.6002
Epoch 4/100
15/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 0.7586 - loss: 0.5206
Epoch 4: val_accuracy improved from 0.68421 to 0.72368, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 2s 100ms/step - accuracy: 0.7617 - loss: 0.5165 - val_accuracy: 0.7237 - val_loss: 0.5158
Epoch 5/100
11/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 0.8588 - loss: 0.3746
Epoch 5: val_accuracy improved from 0.72368 to 0.73684, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 2s 99ms/step - accuracy: 0.8509 - loss: 0.3781 - val_accuracy: 0.7368 - val_loss: 0.4971
Epoch 6/100
11/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 0.8636 - loss: 0.3280
Epoch 6: val_accuracy did not improve from 0.73684
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - accuracy: 0.8663 - loss: 0.3225 - val_accuracy: 0.7237 - val_loss: 0.4887
Epoch 7/100
14/16 ━━━━━━━━━━━━━━━━━━━━ 0s 13ms/step - accuracy: 0.9515 - loss: 0.2058
Epoch 7: val_accuracy improved from 0.73684 to 0.76316, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 2s 108ms/step - accuracy: 0.9488 - loss: 0.2074 - val_accuracy: 0.7632 - val_loss: 0.4961
Epoch 8/100
15/16 ━━━━━━━━━━━━━━━━━━━━ 0s 12ms/step - accuracy: 0.9493 - loss: 0.1521
Epoch 8: val_accuracy improved from 0.76316 to 0.78947, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 2s 97ms/step - accuracy: 0.9484 - loss: 0.1528 - val_accuracy: 0.7895 - val_loss: 0.4980
Epoch 9/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - accuracy: 0.9658 - loss: 0.1137
Epoch 9: val_accuracy improved from 0.78947 to 0.80263, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 2s 94ms/step - accuracy: 0.9659 - loss: 0.1136 - val_accuracy: 0.8026 - val_loss: 0.4939
Epoch 10/100
14/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 0.9620 - loss: 0.1061
Epoch 10: val_accuracy did not improve from 0.80263
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - accuracy: 0.9619 - loss: 0.1061 - val_accuracy: 0.7895 - val_loss: 0.5098
Epoch 11/100
13/16 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - accuracy: 0.9791 - loss: 0.0875
Epoch 11: val_accuracy did not improve from 0.80263
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 29ms/step - accuracy: 0.9779 - loss: 0.0877 - val_accuracy: 0.7895 - val_loss: 0.5424
Epoch 12/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 0.9833 - loss: 0.0730
Epoch 12: val_accuracy improved from 0.80263 to 0.82895, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 2s 105ms/step - accuracy: 0.9836 - loss: 0.0724 - val_accuracy: 0.8289 - val_loss: 0.4979
Epoch 13/100
11/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 0.9913 - loss: 0.0438
Epoch 13: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - accuracy: 0.9906 - loss: 0.0452 - val_accuracy: 0.8289 - val_loss: 0.5162
Epoch 14/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - accuracy: 0.9919 - loss: 0.0446
Epoch 14: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - accuracy: 0.9920 - loss: 0.0443 - val_accuracy: 0.8158 - val_loss: 0.5455
Epoch 15/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 0.9955 - loss: 0.0286
Epoch 15: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - accuracy: 0.9954 - loss: 0.0288 - val_accuracy: 0.8289 - val_loss: 0.5386
Epoch 16/100
12/16 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - accuracy: 0.9991 - loss: 0.0185
Epoch 16: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 38ms/step - accuracy: 0.9974 - loss: 0.0214 - val_accuracy: 0.8026 - val_loss: 0.5602
Epoch 17/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 0.9991 - loss: 0.0258
Epoch 17: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - accuracy: 0.9990 - loss: 0.0258 - val_accuracy: 0.8158 - val_loss: 0.5639
Epoch 18/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 0.9999 - loss: 0.0190
Epoch 18: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - accuracy: 0.9998 - loss: 0.0192 - val_accuracy: 0.8158 - val_loss: 0.5702
Epoch 19/100
13/16 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - accuracy: 0.9982 - loss: 0.0244
Epoch 19: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 31ms/step - accuracy: 0.9981 - loss: 0.0233 - val_accuracy: 0.8289 - val_loss: 0.5724
Epoch 20/100
14/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 1.0000 - loss: 0.0158
Epoch 20: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - accuracy: 1.0000 - loss: 0.0160 - val_accuracy: 0.8158 - val_loss: 0.5862
Epoch 21/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - accuracy: 1.0000 - loss: 0.0182
Epoch 21: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - accuracy: 1.0000 - loss: 0.0181 - val_accuracy: 0.8158 - val_loss: 0.5938
Epoch 22/100
15/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 1.0000 - loss: 0.0154
Epoch 22: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - accuracy: 1.0000 - loss: 0.0154 - val_accuracy: 0.8158 - val_loss: 0.5988
Epoch 23/100
11/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 1.0000 - loss: 0.0159
Epoch 23: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - accuracy: 1.0000 - loss: 0.0153 - val_accuracy: 0.8158 - val_loss: 0.6073
Epoch 24/100
11/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 1.0000 - loss: 0.0132
Epoch 24: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - accuracy: 1.0000 - loss: 0.0132 - val_accuracy: 0.8158 - val_loss: 0.6115
Epoch 25/100
12/16 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - accuracy: 1.0000 - loss: 0.0109
Epoch 25: val_accuracy did not improve from 0.82895
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - accuracy: 1.0000 - loss: 0.0111 - val_accuracy: 0.8289 - val_loss: 0.6086
Epoch 26/100
14/16 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - accuracy: 1.0000 - loss: 0.0098
Epoch 26: val_accuracy improved from 0.82895 to 0.84211, saving model to best_model.weights.h5
16/16 ━━━━━━━━━━━━━━━━━━━━ 2s 111ms/step - accuracy: 1.0000 - loss: 0.0101 - val_accuracy: 0.8421 - val_loss: 0.6128
Epoch 27/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - accuracy: 1.0000 - loss: 0.0136
Epoch 27: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - accuracy: 1.0000 - loss: 0.0134 - val_accuracy: 0.8289 - val_loss: 0.6071
Epoch 28/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - accuracy: 1.0000 - loss: 0.0095
Epoch 28: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - accuracy: 1.0000 - loss: 0.0095 - val_accuracy: 0.8158 - val_loss: 0.6175
Epoch 29/100
15/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 1.0000 - loss: 0.0108
Epoch 29: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - accuracy: 1.0000 - loss: 0.0107 - val_accuracy: 0.8158 - val_loss: 0.6151
Epoch 30/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 1.0000 - loss: 0.0084
Epoch 30: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 33ms/step - accuracy: 1.0000 - loss: 0.0085 - val_accuracy: 0.8289 - val_loss: 0.6145
Epoch 31/100
11/16 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - accuracy: 1.0000 - loss: 0.0082
Epoch 31: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - accuracy: 1.0000 - loss: 0.0083 - val_accuracy: 0.8289 - val_loss: 0.6181
Epoch 32/100
13/16 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - accuracy: 1.0000 - loss: 0.0067
Epoch 32: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 31ms/step - accuracy: 1.0000 - loss: 0.0072 - val_accuracy: 0.8289 - val_loss: 0.6211
Epoch 33/100
15/16 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - accuracy: 1.0000 - loss: 0.0091
Epoch 33: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 35ms/step - accuracy: 1.0000 - loss: 0.0091 - val_accuracy: 0.8289 - val_loss: 0.6219
Epoch 34/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 1.0000 - loss: 0.0068
Epoch 34: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - accuracy: 1.0000 - loss: 0.0069 - val_accuracy: 0.8158 - val_loss: 0.6241
Epoch 35/100
15/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 1.0000 - loss: 0.0066
Epoch 35: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - accuracy: 1.0000 - loss: 0.0068 - val_accuracy: 0.8158 - val_loss: 0.6247
Epoch 36/100
15/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 1.0000 - loss: 0.0082
Epoch 36: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - accuracy: 1.0000 - loss: 0.0082 - val_accuracy: 0.8158 - val_loss: 0.6237
Epoch 37/100
13/16 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - accuracy: 1.0000 - loss: 0.0083
Epoch 37: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 36ms/step - accuracy: 1.0000 - loss: 0.0083 - val_accuracy: 0.8289 - val_loss: 0.6238
Epoch 38/100
11/16 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - accuracy: 1.0000 - loss: 0.0070
Epoch 38: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - accuracy: 1.0000 - loss: 0.0076 - val_accuracy: 0.8289 - val_loss: 0.6234
Epoch 39/100
15/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 1.0000 - loss: 0.0122
Epoch 39: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - accuracy: 1.0000 - loss: 0.0117 - val_accuracy: 0.8289 - val_loss: 0.6225
Epoch 40/100
11/16 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - accuracy: 1.0000 - loss: 0.0053
Epoch 40: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - accuracy: 1.0000 - loss: 0.0062 - val_accuracy: 0.8289 - val_loss: 0.6235
Epoch 41/100
15/16 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - accuracy: 1.0000 - loss: 0.0125
Epoch 41: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - accuracy: 1.0000 - loss: 0.0120 - val_accuracy: 0.8289 - val_loss: 0.6246
Epoch 42/100
12/16 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - accuracy: 1.0000 - loss: 0.0072
Epoch 42: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - accuracy: 1.0000 - loss: 0.0075 - val_accuracy: 0.8289 - val_loss: 0.6251
Epoch 43/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 15ms/step - accuracy: 1.0000 - loss: 0.0084
Epoch 43: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 37ms/step - accuracy: 1.0000 - loss: 0.0084 - val_accuracy: 0.8289 - val_loss: 0.6240
Epoch 44/100
16/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 1.0000 - loss: 0.0067
Epoch 44: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - accuracy: 1.0000 - loss: 0.0068 - val_accuracy: 0.8289 - val_loss: 0.6241
Epoch 45/100
12/16 ━━━━━━━━━━━━━━━━━━━━ 0s 16ms/step - accuracy: 1.0000 - loss: 0.0054
Epoch 45: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - accuracy: 1.0000 - loss: 0.0061 - val_accuracy: 0.8289 - val_loss: 0.6241
Epoch 46/100
11/16 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - accuracy: 1.0000 - loss: 0.0097
Epoch 46: val_accuracy did not improve from 0.84211
16/16 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - accuracy: 1.0000 - loss: 0.0096 - val_accuracy: 0.8289 - val_loss: 0.6244
Epoch 46: 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()

2. 图片预测

# 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("./data/test/nike/1.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) #/255.0  # 记得做归一化处理（与训练集处理方式保持一致）

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)])

1/1 ━━━━━━━━━━━━━━━━━━━━ 1s 775ms/step
预测结果为： adidas

六、总结

• 模型中池化层使用MaxPooling的效果比AveragePooling好
• 模型中使用一次dropout的效果比使用两次好