第T4周：猴痘病识别我的环境操作系统：CentOS7 显卡：RTX3090 两张显卡驱动：550.78 CUDA版本

🍨 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

2024-09-25 14:44:24.624508: I tensorflow/core/util/port.cc:153] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
2024-09-25 14:44:24.642933: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:485] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
2024-09-25 14:44:24.665255: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:8454] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
2024-09-25 14:44:24.671778: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1452] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
2024-09-25 14:44:24.688971: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 AVX512F AVX512_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-09-25 14:44:25.596686: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT





[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU'),
 PhysicalDevice(name='/physical_device:GPU:1', device_type='GPU')]

2.导入查看数据

data_dir = "./data/"

data_dir = pathlib.Path(data_dir)

image_count = len(list(data_dir.glob('*/*.jpg')))

print("The number of pictures",image_count)

The number of pictures 2142

Monkeypox = list(data_dir.glob('Monkeypox/*.jpg'))
PIL.Image.open(str(Monkeypox[100]))

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

"""
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=0.2,
    subset="training",
    seed=123,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 2142 files belonging to 2 classes.
Using 1714 files for training.


2024-09-25 14:44:28.768811: I tensorflow/core/common_runtime/gpu/gpu_device.cc:2021] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 22456 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_dir,
    validation_split=0.2,
    subset="validation",
    seed=123,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 2142 files belonging to 2 classes.
Using 428 files for validation.

class_names = train_ds.class_names
print(class_names)

['Monkeypox', 'Others']

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")

2024-09-25 14:44:30.970847: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence

3.检查数据

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

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

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

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。

网络结构图：

num_classes = 2

"""
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。
在上一篇文章花朵识别中，训练准确率与验证准确率相差巨大就是由于模型过拟合导致的

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.Conv2D(32, (3, 3), activation='relu'),  # 卷积层2，卷积核3*
    layers.AveragePooling2D((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(num_classes)               # 输出层，输出预期结果
])

model.summary()  # 打印网络结构

/project-whj/wangms/bin/miniconda3/envs/CremeNN/lib/python3.9/site-packages/keras/src/layers/preprocessing/tf_data_layer.py:19: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(**kwargs)
/project-whj/wangms/bin/miniconda3/envs/CremeNN/lib/python3.9/site-packages/keras/src/layers/convolutional/base_conv.py:107: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(activity_regularizer=activity_regularizer, **kwargs)

Model: "sequential_4"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ rescaling_4 (Rescaling)         │ (None, 224, 224, 3)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_12 (Conv2D)              │ (None, 222, 222, 16)   │           448 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ average_pooling2d_8             │ (None, 111, 111, 16)   │             0 │
│ (AveragePooling2D)              │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_13 (Conv2D)              │ (None, 109, 109, 32)   │         4,640 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ average_pooling2d_9             │ (None, 54, 54, 32)     │             0 │
│ (AveragePooling2D)              │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout_8 (Dropout)             │ (None, 54, 54, 32)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_14 (Conv2D)              │ (None, 52, 52, 64)     │        18,496 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout_9 (Dropout)             │ (None, 52, 52, 64)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ flatten_4 (Flatten)             │ (None, 173056)         │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_8 (Dense)                 │ (None, 128)            │    22,151,296 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_9 (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)

# 设置优化器
opt = tf.keras.optimizers.Adam(learning_rate=1e-4)

model.compile(optimizer=opt,
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

四、训练模型

from tensorflow.keras.callbacks import ModelCheckpoint

epochs = 50

checkpointer = ModelCheckpoint('best_model.weights.h5',
                                monitor='val_accuracy',
                                verbose=1,
                                save_best_only=True,
                                save_weights_only=True)

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

Epoch 1/50


WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1727247208.518977  258728 service.cc:146] XLA service 0x7fcce4001e60 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
I0000 00:00:1727247208.519039  258728 service.cc:154]   StreamExecutor device (0): NVIDIA GeForce RTX 3090, Compute Capability 8.6
2024-09-25 14:53:28.594313: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:268] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.
2024-09-25 14:53:28.887008: I external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:531] Loaded cuDNN version 8907


[1m16/54[0m [32m━━━━━[0m[37m━━━━━━━━━━━━━━━[0m [1m0s[0m 10ms/step - accuracy: 0.5138 - loss: 0.7708

I0000 00:00:1727247213.916010  258728 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.


[1m53/54[0m [32m━━━━━━━━━━━━━━━━━━━[0m[37m━[0m [1m0s[0m 145ms/step - accuracy: 0.5140 - loss: 0.7350
Epoch 1: val_accuracy improved from -inf to 0.59112, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m19s[0m 212ms/step - accuracy: 0.5148 - loss: 0.7339 - val_accuracy: 0.5911 - val_loss: 0.6614
Epoch 2/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.6250 - loss: 0.6554
Epoch 2: val_accuracy improved from 0.59112 to 0.60280, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 18ms/step - accuracy: 0.6263 - loss: 0.6546 - val_accuracy: 0.6028 - val_loss: 0.6681
Epoch 3/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.6637 - loss: 0.6200
Epoch 3: val_accuracy did not improve from 0.60280
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.6625 - loss: 0.6203 - val_accuracy: 0.6005 - val_loss: 0.7066
Epoch 4/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.6824 - loss: 0.6020
Epoch 4: val_accuracy improved from 0.60280 to 0.60514, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 21ms/step - accuracy: 0.6826 - loss: 0.6012 - val_accuracy: 0.6051 - val_loss: 0.7324
Epoch 5/50
[1m48/54[0m [32m━━━━━━━━━━━━━━━━━[0m[37m━━━[0m [1m0s[0m 8ms/step - accuracy: 0.7008 - loss: 0.5889
Epoch 5: val_accuracy improved from 0.60514 to 0.66589, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.6993 - loss: 0.5890 - val_accuracy: 0.6659 - val_loss: 0.6183
Epoch 6/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.7220 - loss: 0.5590
Epoch 6: val_accuracy improved from 0.66589 to 0.69860, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.7229 - loss: 0.5575 - val_accuracy: 0.6986 - val_loss: 0.5802
Epoch 7/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.7466 - loss: 0.5166
Epoch 7: val_accuracy improved from 0.69860 to 0.74533, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.7460 - loss: 0.5173 - val_accuracy: 0.7453 - val_loss: 0.5016
Epoch 8/50
[1m49/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.7414 - loss: 0.5221
Epoch 8: val_accuracy did not improve from 0.74533
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.7416 - loss: 0.5211 - val_accuracy: 0.7173 - val_loss: 0.5840
Epoch 9/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.7641 - loss: 0.4819
Epoch 9: val_accuracy did not improve from 0.74533
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.7650 - loss: 0.4814 - val_accuracy: 0.7407 - val_loss: 0.4971
Epoch 10/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.7703 - loss: 0.4714
Epoch 10: val_accuracy improved from 0.74533 to 0.77336, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.7724 - loss: 0.4694 - val_accuracy: 0.7734 - val_loss: 0.4536
Epoch 11/50
[1m49/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.8085 - loss: 0.4296
Epoch 11: val_accuracy did not improve from 0.77336
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.8103 - loss: 0.4279 - val_accuracy: 0.7710 - val_loss: 0.4615
Epoch 12/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.8161 - loss: 0.4069
Epoch 12: val_accuracy improved from 0.77336 to 0.80841, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.8176 - loss: 0.4056 - val_accuracy: 0.8084 - val_loss: 0.4040
Epoch 13/50
[1m49/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.8642 - loss: 0.3447
Epoch 13: val_accuracy improved from 0.80841 to 0.81308, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.8620 - loss: 0.3467 - val_accuracy: 0.8131 - val_loss: 0.4125
Epoch 14/50
[1m52/54[0m [32m━━━━━━━━━━━━━━━━━━━[0m[37m━[0m [1m0s[0m 8ms/step - accuracy: 0.8653 - loss: 0.3479
Epoch 14: val_accuracy improved from 0.81308 to 0.82944, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 23ms/step - accuracy: 0.8647 - loss: 0.3485 - val_accuracy: 0.8294 - val_loss: 0.3971
Epoch 15/50
[1m53/54[0m [32m━━━━━━━━━━━━━━━━━━━[0m[37m━[0m [1m0s[0m 8ms/step - accuracy: 0.8694 - loss: 0.3446
Epoch 15: val_accuracy improved from 0.82944 to 0.84112, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 20ms/step - accuracy: 0.8693 - loss: 0.3443 - val_accuracy: 0.8411 - val_loss: 0.3976
Epoch 16/50
[1m53/54[0m [32m━━━━━━━━━━━━━━━━━━━[0m[37m━[0m [1m0s[0m 8ms/step - accuracy: 0.8733 - loss: 0.3238
Epoch 16: val_accuracy did not improve from 0.84112
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.8736 - loss: 0.3233 - val_accuracy: 0.8411 - val_loss: 0.3665
Epoch 17/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.8690 - loss: 0.3047
Epoch 17: val_accuracy did not improve from 0.84112
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.8683 - loss: 0.3061 - val_accuracy: 0.8271 - val_loss: 0.3639
Epoch 18/50
[1m49/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.8898 - loss: 0.2766
Epoch 18: val_accuracy did not improve from 0.84112
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.8887 - loss: 0.2785 - val_accuracy: 0.8224 - val_loss: 0.3781
Epoch 19/50
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 8ms/step - accuracy: 0.8850 - loss: 0.2790
Epoch 19: val_accuracy did not improve from 0.84112
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.8851 - loss: 0.2790 - val_accuracy: 0.8154 - val_loss: 0.4185
Epoch 20/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.8954 - loss: 0.2759
Epoch 20: val_accuracy improved from 0.84112 to 0.86215, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 20ms/step - accuracy: 0.8953 - loss: 0.2753 - val_accuracy: 0.8621 - val_loss: 0.3685
Epoch 21/50
[1m49/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9174 - loss: 0.2414
Epoch 21: val_accuracy did not improve from 0.86215
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9169 - loss: 0.2414 - val_accuracy: 0.8248 - val_loss: 0.4002
Epoch 22/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.8827 - loss: 0.2790
Epoch 22: val_accuracy did not improve from 0.86215
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.8834 - loss: 0.2781 - val_accuracy: 0.8551 - val_loss: 0.3512
Epoch 23/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9192 - loss: 0.2118
Epoch 23: val_accuracy did not improve from 0.86215
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9182 - loss: 0.2137 - val_accuracy: 0.8528 - val_loss: 0.3585
Epoch 24/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9200 - loss: 0.2052
Epoch 24: val_accuracy did not improve from 0.86215
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9202 - loss: 0.2047 - val_accuracy: 0.8294 - val_loss: 0.4180
Epoch 25/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9051 - loss: 0.2240
Epoch 25: val_accuracy improved from 0.86215 to 0.86916, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.9064 - loss: 0.2226 - val_accuracy: 0.8692 - val_loss: 0.3502
Epoch 26/50
[1m48/54[0m [32m━━━━━━━━━━━━━━━━━[0m[37m━━━[0m [1m0s[0m 8ms/step - accuracy: 0.9359 - loss: 0.1884
Epoch 26: val_accuracy did not improve from 0.86916
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9347 - loss: 0.1897 - val_accuracy: 0.8458 - val_loss: 0.3741
Epoch 27/50
[1m49/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9363 - loss: 0.1863
Epoch 27: val_accuracy did not improve from 0.86916
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9359 - loss: 0.1857 - val_accuracy: 0.8692 - val_loss: 0.3452
Epoch 28/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9377 - loss: 0.1608
Epoch 28: val_accuracy did not improve from 0.86916
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9373 - loss: 0.1615 - val_accuracy: 0.8621 - val_loss: 0.3532
Epoch 29/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9449 - loss: 0.1650
Epoch 29: val_accuracy did not improve from 0.86916
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9443 - loss: 0.1655 - val_accuracy: 0.8364 - val_loss: 0.4030
Epoch 30/50
[1m53/54[0m [32m━━━━━━━━━━━━━━━━━━━[0m[37m━[0m [1m0s[0m 8ms/step - accuracy: 0.9436 - loss: 0.1570
Epoch 30: val_accuracy did not improve from 0.86916
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9434 - loss: 0.1576 - val_accuracy: 0.8551 - val_loss: 0.3577
Epoch 31/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9364 - loss: 0.1640
Epoch 31: val_accuracy improved from 0.86916 to 0.87150, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.9373 - loss: 0.1627 - val_accuracy: 0.8715 - val_loss: 0.3598
Epoch 32/50
[1m53/54[0m [32m━━━━━━━━━━━━━━━━━━━[0m[37m━[0m [1m0s[0m 8ms/step - accuracy: 0.9605 - loss: 0.1308
Epoch 32: val_accuracy did not improve from 0.87150
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9602 - loss: 0.1314 - val_accuracy: 0.8481 - val_loss: 0.4216
Epoch 33/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9343 - loss: 0.1684
Epoch 33: val_accuracy did not improve from 0.87150
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9348 - loss: 0.1678 - val_accuracy: 0.8435 - val_loss: 0.4420
Epoch 34/50
[1m53/54[0m [32m━━━━━━━━━━━━━━━━━━━[0m[37m━[0m [1m0s[0m 8ms/step - accuracy: 0.9492 - loss: 0.1350
Epoch 34: val_accuracy improved from 0.87150 to 0.87383, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.9492 - loss: 0.1348 - val_accuracy: 0.8738 - val_loss: 0.3825
Epoch 35/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9659 - loss: 0.1127
Epoch 35: val_accuracy did not improve from 0.87383
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9656 - loss: 0.1126 - val_accuracy: 0.8738 - val_loss: 0.3742
Epoch 36/50
[1m49/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9627 - loss: 0.1115
Epoch 36: val_accuracy did not improve from 0.87383
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9627 - loss: 0.1109 - val_accuracy: 0.8738 - val_loss: 0.3912
Epoch 37/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9673 - loss: 0.1002
Epoch 37: val_accuracy did not improve from 0.87383
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9668 - loss: 0.1006 - val_accuracy: 0.8738 - val_loss: 0.3848
Epoch 38/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9774 - loss: 0.0920
Epoch 38: val_accuracy improved from 0.87383 to 0.87850, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.9772 - loss: 0.0921 - val_accuracy: 0.8785 - val_loss: 0.3904
Epoch 39/50
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m0s[0m 8ms/step - accuracy: 0.9644 - loss: 0.1004
Epoch 39: val_accuracy did not improve from 0.87850
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9642 - loss: 0.1007 - val_accuracy: 0.8505 - val_loss: 0.4173
Epoch 40/50
[1m48/54[0m [32m━━━━━━━━━━━━━━━━━[0m[37m━━━[0m [1m0s[0m 8ms/step - accuracy: 0.9628 - loss: 0.1003
Epoch 40: val_accuracy did not improve from 0.87850
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9637 - loss: 0.0993 - val_accuracy: 0.8715 - val_loss: 0.3890
Epoch 41/50
[1m52/54[0m [32m━━━━━━━━━━━━━━━━━━━[0m[37m━[0m [1m0s[0m 8ms/step - accuracy: 0.9850 - loss: 0.0729
Epoch 41: val_accuracy did not improve from 0.87850
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9845 - loss: 0.0736 - val_accuracy: 0.8668 - val_loss: 0.3961
Epoch 42/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9796 - loss: 0.0735
Epoch 42: val_accuracy did not improve from 0.87850
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9794 - loss: 0.0733 - val_accuracy: 0.8738 - val_loss: 0.4208
Epoch 43/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9813 - loss: 0.0743
Epoch 43: val_accuracy improved from 0.87850 to 0.88084, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 18ms/step - accuracy: 0.9809 - loss: 0.0746 - val_accuracy: 0.8808 - val_loss: 0.4223
Epoch 44/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9769 - loss: 0.0671
Epoch 44: val_accuracy improved from 0.88084 to 0.88551, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 18ms/step - accuracy: 0.9767 - loss: 0.0678 - val_accuracy: 0.8855 - val_loss: 0.4224
Epoch 45/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9613 - loss: 0.0881
Epoch 45: val_accuracy did not improve from 0.88551
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9624 - loss: 0.0868 - val_accuracy: 0.8832 - val_loss: 0.4132
Epoch 46/50
[1m52/54[0m [32m━━━━━━━━━━━━━━━━━━━[0m[37m━[0m [1m0s[0m 8ms/step - accuracy: 0.9802 - loss: 0.0659
Epoch 46: val_accuracy did not improve from 0.88551
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9802 - loss: 0.0660 - val_accuracy: 0.8738 - val_loss: 0.4336
Epoch 47/50
[1m48/54[0m [32m━━━━━━━━━━━━━━━━━[0m[37m━━━[0m [1m0s[0m 8ms/step - accuracy: 0.9809 - loss: 0.0646
Epoch 47: val_accuracy improved from 0.88551 to 0.89019, saving model to best_model.weights.h5
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 19ms/step - accuracy: 0.9809 - loss: 0.0650 - val_accuracy: 0.8902 - val_loss: 0.4047
Epoch 48/50
[1m49/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9848 - loss: 0.0621
Epoch 48: val_accuracy did not improve from 0.89019
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9848 - loss: 0.0615 - val_accuracy: 0.8621 - val_loss: 0.5090
Epoch 49/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 8ms/step - accuracy: 0.9797 - loss: 0.0596
Epoch 49: val_accuracy did not improve from 0.89019
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 9ms/step - accuracy: 0.9797 - loss: 0.0599 - val_accuracy: 0.8785 - val_loss: 0.4285
Epoch 50/50
[1m50/54[0m [32m━━━━━━━━━━━━━━━━━━[0m[37m━━[0m [1m0s[0m 9ms/step - accuracy: 0.9901 - loss: 0.0465
Epoch 50: val_accuracy did not improve from 0.89019
[1m54/54[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 10ms/step - accuracy: 0.9893 - loss: 0.0479 - val_accuracy: 0.8505 - val_loss: 0.4994

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(epochs)

plt.figure(figsize=(12, 4))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, acc, label='Training Accuracy')
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')

plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()

2. 指定图片进行预测

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

from PIL import Image
import numpy as np

# img = Image.open("./45-data/Monkeypox/M06_01_04.jpg")  #这里选择你需要预测的图片
img = Image.open("./data/Others/NM15_02_11.jpg")  #这里选择你需要预测的图片
image = tf.image.resize(img, [img_height, img_width])

img_array = tf.expand_dims(image, 0) 

predictions = model.predict(img_array) # 这里选用你已经训练好的模型
print("预测结果为：",class_names[np.argmax(predictions)])

[1m1/1[0m [32m━━━━━━━━━━━━━━━━━━━━[0m[37m[0m [1m1s[0m 673ms/step
预测结果为： Others


img = Image.open("./data/Monkeypox/M06_01_04.jpg")  #这里选择你需要预测的图片
image = tf.image.resize(img, [img_height, img_width])

img_array = tf.expand_dims(image, 0) 

predictions = model.predict(img_array) # 这里选用你已经训练好的模型
print("预测结果为：",class_names[np.argmax(predictions)])

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

六、总结

老版本的 Tensorflow 的 layers.experimental.preprocessing.Rescaling 方法在新版本中简化为 layers.Rescaling