深度学习训练营之数据增强

news2024/11/27 15:39:58

深度学习训练营

学习内容
原文链接
环境介绍
前置工作
- 设置GPU
- 加载数据
- 创建测试集
- 数据类型查看以及数据归一化
数据增强操作
- 使用嵌入model的方法进行数据增强
模型训练
结果可视化
- 自定义数据增强
查看数据增强后的图片

学习内容

在深度学习当中,由于准备数据集本身是一件十分复杂的过程,很难保障每一张图片的学习能力都很高,所以对于同一种图片采用数据增强就显得十分重要了

原文链接

🍨 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P10周:实现数据增强
🍖 原作者：K同学啊|接辅导、项目定制

环境介绍

语言环境：Python3.9.13
编译器：jupyter notebook
深度学习环境：TensorFlow2
数据链接:猫和狗数据

前置工作

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)

加载数据

将对应的数据按照不同种类放入到不同文件夹当中,再将数据整合为animal_data

data_dir   = "animal_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.

创建测试集

因为数据本身没有设置测试集,这里需要进行手动创建

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))

运行结构如下

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

预测的batches和测试batches分别为60和15

数据类型查看以及数据归一化

class_names = train_ds.class_names
print(class_names)

['cat', 'dog']

进行数据归一化操作

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)

查看数据集

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")

在这里插入图片描述

数据增强操作

data_augmentation = tf.keras.Sequential([
  tf.keras.layers.experimental.preprocessing.RandomFlip("horizontal_and_vertical"),
  tf.keras.layers.experimental.preprocessing.RandomRotation(0.2),
])
#进行随机的水平翻转和垂直翻转
# 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")

请添加图片描述

使用嵌入model的方法进行数据增强

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

这样的操作可以得到GPU的加速

模型训练

模型开始训练之前都需要进行这个模型的调整

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'])

开始进行正式的训练

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

75/75 [==============================] - 35s 462ms/step - loss: 2.0268e-05 - accuracy: 1.0000 - val_loss: 1.8425e-05 - val_accuracy: 1.0000
Epoch 2/20
75/75 [==============================] - 34s 461ms/step - loss: 1.7937e-05 - accuracy: 1.0000 - val_loss: 1.6272e-05 - val_accuracy: 1.0000
Epoch 3/20
75/75 [==============================] - 35s 461ms/step - loss: 1.5871e-05 - accuracy: 1.0000 - val_loss: 1.4373e-05 - val_accuracy: 1.0000
Epoch 4/20
75/75 [==============================] - 34s 450ms/step - loss: 1.4039e-05 - accuracy: 1.0000 - val_loss: 1.2682e-05 - val_accuracy: 1.0000
Epoch 5/20
75/75 [==============================] - 34s 450ms/step - loss: 1.2429e-05 - accuracy: 1.0000 - val_loss: 1.1195e-05 - val_accuracy: 1.0000
Epoch 6/20
75/75 [==============================] - 35s 462ms/step - loss: 1.1014e-05 - accuracy: 1.0000 - val_loss: 9.8961e-06 - val_accuracy: 1.0000
Epoch 7/20
75/75 [==============================] - 34s 450ms/step - loss: 9.7220e-06 - accuracy: 1.0000 - val_loss: 8.6961e-06 - val_accuracy: 1.0000
Epoch 8/20
75/75 [==============================] - 34s 455ms/step - loss: 8.5416e-06 - accuracy: 1.0000 - val_loss: 7.6252e-06 - val_accuracy: 1.0000
Epoch 9/20
75/75 [==============================] - 34s 459ms/step - loss: 7.5130e-06 - accuracy: 1.0000 - val_loss: 6.7169e-06 - val_accuracy: 1.0000
Epoch 10/20
75/75 [==============================] - 34s 460ms/step - loss: 6.6338e-06 - accuracy: 1.0000 - val_loss: 5.9490e-06 - val_accuracy: 1.0000
Epoch 11/20
75/75 [==============================] - 34s 457ms/step - loss: 5.8835e-06 - accuracy: 1.0000 - val_loss: 5.2946e-06 - val_accuracy: 1.0000
Epoch 12/20
75/75 [==============================] - 34s 456ms/step - loss: 5.2507e-06 - accuracy: 1.0000 - val_loss: 4.7294e-06 - val_accuracy: 1.0000
Epoch 13/20
...
Epoch 19/20
75/75 [==============================] - 34s 449ms/step - loss: 2.5978e-06 - accuracy: 1.0000 - val_loss: 2.3737e-06 - val_accuracy: 1.0000
Epoch 20/20
75/75 [==============================] - 34s 449ms/step - loss: 2.3849e-06 - accuracy: 1.0000 - val_loss: 2.1841e-06 - val_accuracy: 1.0000

这里比较奇怪的是训练的结果准确性很高,loss的值都是很小很小的,和原本博主的相应的内容是不一样的,我觉得很大的可能应该是首先这个数据的内容很大,原本只有几百张图片,但是这里一共有3400张图片,再加上模型训练的增强方式比较简单,导致在结果上面训练看起来很好

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

15/15 [==============================] - 1s 83ms/step - loss: 1.9960e-06 - accuracy: 1.0000
Accuracy 1.0

结果可视化

自定义数据增强

这里主要是可以更改随机数种子的大小

import random
# 这是大家可以自由发挥的一个地方
def aug_img(image):
    seed = (random.randint(5,10), 0)
    #设立随机数种植,randint是指在0到9之间进行一个数据的增强
    # 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

查看数据增强后的图片

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")