第T10周：数据增强

• 🍨 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 | 接辅导、项目定制

一、前期工作

1.设置GPU

import matplotlib.pyplot as plt
import numpy as np
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)
    tf.config.set_visible_devices([gpus[0]],"GPU")

print(gpus)

输出
[PhysicalDevice(name=‘/physical_device:GPU:0’, device_type=‘GPU’)]

1.加载数据

data_dir   = "./P8/"
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)

输出
Found 3403 files belonging to 2 classes.
Using 2383 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)

输出
Found 3403 files belonging to 2 classes.
Using 2383 files for training.

train_ds = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir,
    validation_split=0.2,
    subset="training",
    seed=12,
    image_size=(img_height, img_width),
    batch_size=batch_size)

输出
Found 3403 files belonging to 2 classes.
Using 2723 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))

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

class_names = train_ds.class_names
print(class_names)

输出
[‘cat’, ‘dog’]

AUTOTUNE = tf.data.AUTOTUNE

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

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))  

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")

输出

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),
])

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")

输出

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

train_ds = prepare(train_ds)

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
)

输出
5/75 [] - 17s 38ms/step - loss: 0.6229 - accuracy: 0.7423 - val_loss: 0.1964 - val_accuracy: 0.9275
Epoch 2/20
75/75 [] - 2s 22ms/step - loss: 0.1831 - accuracy: 0.9324 - val_loss: 0.1277 - val_accuracy: 0.9559
Epoch 3/20
75/75 [] - 2s 22ms/step - loss: 0.0841 - accuracy: 0.9727 - val_loss: 0.0423 - val_accuracy: 0.9842
Epoch 4/20
75/75 [] - 2s 21ms/step - loss: 0.0648 - accuracy: 0.9790 - val_loss: 0.0305 - val_accuracy: 0.9895
Epoch 5/20
75/75 [] - 2s 27ms/step - loss: 0.0359 - accuracy: 0.9874 - val_loss: 0.0166 - val_accuracy: 0.9958
Epoch 6/20
75/75 [] - 2s 26ms/step - loss: 0.0197 - accuracy: 0.9937 - val_loss: 0.0044 - val_accuracy: 1.0000
Epoch 7/20
75/75 [] - 2s 25ms/step - loss: 0.0144 - accuracy: 0.9966 - val_loss: 0.0061 - val_accuracy: 0.9968
Epoch 8/20
75/75 [] - 2s 22ms/step - loss: 0.0165 - accuracy: 0.9958 - val_loss: 0.0922 - val_accuracy: 0.9648
Epoch 9/20
75/75 [] - 2s 20ms/step - loss: 0.0133 - accuracy: 0.9954 - val_loss: 0.0076 - val_accuracy: 0.9979
Epoch 10/20
75/75 [] - 2s 23ms/step - loss: 0.0115 - accuracy: 0.9966 - val_loss: 0.0073 - val_accuracy: 0.9989
Epoch 11/20
75/75 [] - 2s 25ms/step - loss: 0.0098 - accuracy: 0.9966 - val_loss: 0.0053 - val_accuracy: 0.9989
Epoch 12/20
75/75 [] - 2s 23ms/step - loss: 0.0087 - accuracy: 0.9966 - val_loss: 0.0042 - val_accuracy: 0.9984
Epoch 13/20
75/75 [] - 2s 22ms/step - loss: 0.0154 - accuracy: 0.9941 - val_loss: 0.0540 - val_accuracy: 0.9863
Epoch 14/20
75/75 [] - 2s 21ms/step - loss: 0.0233 - accuracy: 0.9912 - val_loss: 0.0011 - val_accuracy: 0.9995
Epoch 15/20
75/75 [] - 2s 20ms/step - loss: 0.0015 - accuracy: 0.9996 - val_loss: 1.7385e-04 - val_accuracy: 1.0000
Epoch 16/20
75/75 [] - 2s 24ms/step - loss: 2.0985e-04 - accuracy: 1.0000 - val_loss: 1.0143e-04 - val_accuracy: 1.0000
Epoch 17/20
75/75 [] - 2s 22ms/step - loss: 1.1244e-04 - accuracy: 1.0000 - val_loss: 7.3976e-05 - val_accuracy: 1.0000
Epoch 18/20
75/75 [] - 2s 23ms/step - loss: 8.2623e-05 - accuracy: 1.0000 - val_loss: 5.8787e-05 - val_accuracy: 1.0000
Epoch 19/20
75/75 [] - 2s 23ms/step - loss: 6.6407e-05 - accuracy: 1.0000 - val_loss: 4.8501e-05 - val_accuracy: 1.0000
Epoch 20/20
75/75 [] - 2s 21ms/step - loss: 5.5053e-05 - accuracy: 1.0000 - val_loss: 4.1066e-05 - val_accuracy: 1.0000

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

输出
15/15 [==============================] - 0s 13ms/step - loss: 5.6710e-05 - accuracy: 1.0000
Accuracy 1.0

import random

def aug_img(image):
    seed = (random.randint(0,8), 0)
    stateless_random_brightness = tf.image.stateless_random_contrast(image, lower=0.3, 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())

输出
Min and max pixel values: 0.0 255.0

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")

输出