T10: 数据增强

• 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
• 🍦 参考文章：365天深度学习训练营-第10周：数据增强（训练营内部成员可读）
• 🍖 原作者：K同学啊 | 接辅导、项目定制
• 🚀 文章来源：K同学的学习圈子

Z. 心得感受+知识点补充

1. 数据增强的两种方式：可以使用 tf.keras.layers.experimental.preprocessing.RandomFlip 与 tf.keras.layers.experimental.preprocessing.RandomRotation 进行数据增强

• tf.keras.layers.experimental.preprocessing.RandomFlip：水平和垂直随机翻转每个图像。
• tf.keras.layers.experimental.preprocessing.RandomRotation：随机旋转每个图像。

2. 增强的两种方式：
   方法一：将其嵌入model中

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

这样做的好处是：

• 数据增强这块的工作可以得到GPU的加速（如果你使用了GPU训练的话）
• 注意：只有在模型训练时（Model.fit）才会进行增强，在模型评估(Model.evaluate)以及预测(Model.predict)时并不会进行增强操作。
  方法二：在Dataset数据集中进行数据增强

batch_size = 32
AUTOTUNE = tf.data.AUTOTUNE

def prepare(ds):
    ds = ds.map(lambda x, y: (data_augmentation(x, training=True), y), num_parallel_calls=AUTOTUNE)
    return ds

一、前期工作

1. 设置GPU

import matplotlib.pyplot as plt
import numpy as np
#隐藏警告
import warnings
warnings.filterwarnings('ignore')

from tensorflow.keras import layers
import tensorflow as tf
gpus = tf.config.list_physical_devices("GPU")

if gpus:
    tf.config.experimental.set_memory_growth(gpus[0], True)  #设置GPU显存用量按需使用
    tf.config.set_visible_devices([gpus[0]],"GPU")

# 打印显卡信息，确认GPU可用
print(gpus)

[]

2. 加载数据

data_dir   = "./34-data/"
img_height = 224
img_width  = 224
batch_size = 32

train_ds = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir,
    validation_split=0.3,
    subset="training",
    seed=12,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 3400 files belonging to 2 classes.
Using 2380 files for training.

val_ds = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir,
    validation_split=0.3,
    subset="training",
    seed=12,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 3400 files belonging to 2 classes.
Using 2380 files for training.

val_batches = tf.data.experimental.cardinality(val_ds)
test_ds     = val_ds.take(val_batches // 5)
val_ds      = val_ds.skip(val_batches // 5)

print('Number of validation batches: %d' % tf.data.experimental.cardinality(val_ds))
print('Number of test batches: %d' % tf.data.experimental.cardinality(test_ds))

Number of validation batches: 60
Number of test batches: 15

class_names = train_ds.class_names
print(class_names)

[‘cat’, ‘dog’]

AUTOTUNE = tf.data.AUTOTUNE

def preprocess_image(image,label):
    return (image/255.0,label)

# 归一化处理
train_ds = train_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)
val_ds   = val_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)
test_ds  = test_ds.map(preprocess_image, num_parallel_calls=AUTOTUNE)

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

plt.figure(figsize=(15, 10))  # 图形的宽为15高为10

for images, labels in train_ds.take(1):
    for i in range(8):
        
        ax = plt.subplot(5, 8, i + 1) 
        plt.imshow(images[i])
        plt.title(class_names[labels[i]])
        
        plt.axis("off")

二、数据增强

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

# Add the image to a batch.
image = tf.expand_dims(images[i], 0)

plt.figure(figsize=(8, 8))
for i in range(9):
    augmented_image = data_augmentation(image)
    ax = plt.subplot(3, 3, i + 1)
    plt.imshow(augmented_image[0])
    plt.axis("off")

三、增强方式

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

四、训练模型

model = tf.keras.Sequential([
  layers.Conv2D(16, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(32, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Conv2D(64, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  layers.Flatten(),
  layers.Dense(128, activation='relu'),
  layers.Dense(len(class_names))
])

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

epochs=20
history = model.fit(
  train_ds,
  validation_data=val_ds,
  epochs=epochs
)

Epoch 1/20
75/75 [] - 32s 422ms/step - loss: 0.0866 - accuracy: 0.9693 - val_loss: 0.0434 - val_accuracy: 0.9847
Epoch 2/20
75/75 [] - 32s 427ms/step - loss: 0.0640 - accuracy: 0.9756 - val_loss: 0.0254 - val_accuracy: 0.9926
Epoch 3/20
75/75 [] - 32s 431ms/step - loss: 0.0411 - accuracy: 0.9891 - val_loss: 0.0163 - val_accuracy: 0.9953
Epoch 4/20
75/75 [] - 32s 431ms/step - loss: 0.0197 - accuracy: 0.9937 - val_loss: 0.0030 - val_accuracy: 1.0000
Epoch 5/20
75/75 [] - 35s 472ms/step - loss: 0.0028 - accuracy: 1.0000 - val_loss: 0.0017 - val_accuracy: 1.0000
Epoch 6/20
75/75 [] - 32s 432ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 7.3165e-04 - val_accuracy: 1.0000
Epoch 7/20
75/75 [] - 32s 433ms/step - loss: 8.3814e-04 - accuracy: 1.0000 - val_loss: 4.2073e-04 - val_accuracy: 1.0000
Epoch 8/20
75/75 [] - 32s 433ms/step - loss: 2.9329e-04 - accuracy: 1.0000 - val_loss: 2.1425e-04 - val_accuracy: 1.0000
Epoch 9/20
75/75 [] - 37s 490ms/step - loss: 1.8117e-04 - accuracy: 1.0000 - val_loss: 1.6512e-04 - val_accuracy: 1.0000
Epoch 10/20
75/75 [] - 33s 443ms/step - loss: 1.4339e-04 - accuracy: 1.0000 - val_loss: 1.3430e-04 - val_accuracy: 1.0000
Epoch 11/20
75/75 [] - 34s 459ms/step - loss: 1.1867e-04 - accuracy: 1.0000 - val_loss: 1.1258e-04 - val_accuracy: 1.0000
Epoch 12/20
75/75 [] - 37s 488ms/step - loss: 1.0092e-04 - accuracy: 1.0000 - val_loss: 9.3655e-05 - val_accuracy: 1.0000
Epoch 13/20
75/75 [] - 35s 464ms/step - loss: 8.5331e-05 - accuracy: 1.0000 - val_loss: 7.8893e-05 - val_accuracy: 1.0000
Epoch 14/20
75/75 [] - 35s 468ms/step - loss: 7.2413e-05 - accuracy: 1.0000 - val_loss: 6.8605e-05 - val_accuracy: 1.0000
Epoch 15/20
75/75 [] - 36s 479ms/step - loss: 6.2746e-05 - accuracy: 1.0000 - val_loss: 5.4928e-05 - val_accuracy: 1.0000
Epoch 16/20
75/75 [] - 34s 452ms/step - loss: 5.2526e-05 - accuracy: 1.0000 - val_loss: 4.7366e-05 - val_accuracy: 1.0000
Epoch 17/20
75/75 [] - 36s 476ms/step - loss: 4.4461e-05 - accuracy: 1.0000 - val_loss: 4.1247e-05 - val_accuracy: 1.0000
Epoch 18/20
75/75 [] - 33s 442ms/step - loss: 3.7998e-05 - accuracy: 1.0000 - val_loss: 3.6842e-05 - val_accuracy: 1.0000
Epoch 19/20
75/75 [] - 33s 439ms/step - loss: 3.2455e-05 - accuracy: 1.0000 - val_loss: 3.1903e-05 - val_accuracy: 1.0000
Epoch 20/20
75/75 [] - 33s 443ms/step - loss: 2.8046e-05 - accuracy: 1.0000 - val_loss: 2.7984e-05 - val_accuracy: 1.0000

loss, acc = model.evaluate(test_ds)
print("Accuracy", acc)

15/15 [==============================] - 1s 85ms/step - loss: 2.5530e-05 - accuracy: 1.0000
Accuracy 1.0

五、自定义增强函数

import random
# 这是大家可以自由发挥的一个地方
def aug_img(image):
    seed = (random.randint(0,9), 0)
    # 随机改变图像对比度
    stateless_random_brightness = tf.image.stateless_random_contrast(image, lower=0.1, upper=1.0, seed=seed)
    return stateless_random_brightness

image = tf.expand_dims(images[3]*255, 0)
print("Min and max pixel values:", image.numpy().min(), image.numpy().max())

Min and max pixel values: 2.4591687 241.47968

plt.figure(figsize=(8, 8))
for i in range(9):
    augmented_image = aug_img(image)
    ax = plt.subplot(3, 3, i + 1)
    plt.imshow(augmented_image[0].numpy().astype("uint8"))

    plt.axis("off")