如何在 python 中为我的自定义数据创建流生成器

Question

我做猫/狗二元分类 我以这种方式创建了训练数据，我对图像应用了平均过滤器。 问题是数据库很大，之后我就显示了，你的笔记本试图分配比可用内存更多的内存。我读到 python 中的生成器占用更少的磁盘内存并且可以解决这个问题，但我不知道如何创建一个适合我刚刚创建为训练数据的代码的生成器

train_dir = "../input/dog-cat/train"

CATEGORIES = ["dog", "cat"]
    
training_data = []

def create_training_data():
    for category in CATEGORIES:  

        path = os.path.join(train_dir,category)  
        class_num = CATEGORIES.index(category)  

        for img in tqdm(os.listdir(path)):  
            try:
                img_train = cv2.imread(os.path.join(path,img))
                img_mean = cv2.blur(reduced_img_train,(9,9))
                training_data.append([img_mean, class_num])  
            except Exception as e:
             pass
create_training_data()

import random
random.shuffle(training_data)

x_train=[]
y_train=[]

for features,label in training_data:
    x_train.append(features)
    y_train.append(label)

Answer 1

根据您想要使用带模糊功能的 ImageDataGenerator() 的要求，请查看 CV2 CV2.blur()。您可以通过 ImageDataGenerator() 本身提供的自定义函数“preprocessing_function=custom_image_preprocess”参数来完成。

示例：当您可以使用自定义函数或仅使用相同的图像通道顺序时使用标准偏差的 CV2（跆拳道颜色游戏中可重建数据的一种隐藏技术）。

import tensorflow as tf

import matplotlib.pyplot as plt

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
None
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
physical_devices = tf.config.experimental.list_physical_devices('GPU')
assert len(physical_devices) > 0, "Not enough GPU hardware devices available"
config = tf.config.experimental.set_memory_growth(physical_devices[0], True)
print(physical_devices)
print(config)

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Variables
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
BATCH_SIZE = 1
IMG_HEIGHT = 32
IMG_WIDTH = 32
IMG_CHANNELS=3
seed=42

directory = "F:\datasets\downloads\example\image\"

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Definition / Class
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
def custom_image_preprocess( image ):
    image = tf.keras.preprocessing.image.array_to_img(
        image,
        data_format=None,
        scale=True
    )
    img_array = tf.keras.preprocessing.image.img_to_array( image )
    img_1 = tf.keras.utils.array_to_img(img_array)
    
    temp = tf.concat([ tf.constant( img_array[:,:,0], shape=(img_array.shape[0], img_array.shape[1], 1) ), tf.constant( 150 - img_array[:,:,1], shape=(img_array.shape[0], img_array.shape[1], 1) ) ], axis=2)
    image = tf.concat([ tf.constant( temp[:,:,:], shape=(img_array.shape[0], img_array.shape[1], 2) ), tf.constant( 0.25 * img_array[:,:,2], shape=(img_array.shape[0], img_array.shape[1], 1) ) ], axis=2)

    return image


def train_image_gen():

    n_zoom_range = tf.where( tf.math.greater_equal( tf.constant( ( 1.0 * IMG_WIDTH ) / ( IMG_HEIGHT * 4 ), dtype=tf.float32 ), tf.constant( 0.25, dtype=tf.float32 ) ), ( 1.0 * IMG_WIDTH ) / ( IMG_HEIGHT * 4 ), 0.25 ).numpy()
    n_rotation_range = tf.where( tf.math.greater_equal( tf.constant( ( 1.0 * IMG_WIDTH ) / ( IMG_HEIGHT * 4 ), dtype=tf.float32 ), tf.constant( 0.25, dtype=tf.float32 ) ), ( 1.0 * IMG_WIDTH ) / ( IMG_HEIGHT * 4 ) * 100, 27.25 ).numpy()
    n_rescale = tf.where( tf.math.less_equal( tf.constant( 1.0 / ( IMG_WIDTH + IMG_HEIGHT )), tf.constant( 125.0 )), tf.constant( 1.0 / ( IMG_WIDTH + IMG_HEIGHT )).numpy(), 125.0 ).numpy()

    train_generator = tf.keras.preprocessing.image.ImageDataGenerator(
        # shear_range=0.2,
        # zoom_range=float(n_zoom_range),
        # horizontal_flip=True,
        validation_split=0.2,
        # rotation_range=float(n_rotation_range),
        # rescale=float(n_rescale),
        
        # rescale=1./255,
        # featurewise_center=False,
        # samplewise_center=False,
        # featurewise_std_normalization=False,
        # samplewise_std_normalization=False,
        # zca_whitening=False,
        # zca_epsilon=1e-06,
        # rotation_range=0,
        # width_shift_range=0.0,
        # height_shift_range=0.0,
        # brightness_range=None,
        # shear_range=0.0,
        # zoom_range=0.0,
        # channel_shift_range=0.0,
        # fill_mode='nearest',
        # cval=0.0,
        # horizontal_flip=False,
        # vertical_flip=False,
        # rescale=None,
        preprocessing_function=custom_image_preprocess
        # data_format=None,
        # validation_split=0.0,
        # interpolation_order=1,
        # dtype=None
        # https://www.tensorflow.org/api_docs/python/tf/keras/preprocessing/image/ImageDataGenerator
        
        )
        
    train_image_ds = train_generator.flow_from_directory(
        directory,
        target_size=(IMG_HEIGHT, IMG_WIDTH),
        batch_size=BATCH_SIZE,
        class_mode='binary',    # None  # categorical   # binary
        subset='training',
        color_mode='rgb',       # rgb   # grayscale
        seed=seed,
        )
        
    return train_image_ds

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Model Initialize
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
model = tf.keras.models.Sequential([
    tf.keras.layers.InputLayer(input_shape=( IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS )),
    tf.keras.layers.Reshape((IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS)),
    tf.keras.layers.RandomFlip('horizontal'),
    tf.keras.layers.RandomRotation(0.2),
    tf.keras.layers.Normalization(mean=3., variance=2.),
    tf.keras.layers.Normalization(mean=4., variance=6.),
    tf.keras.layers.Conv2D(32, (3, 3), activation='relu'),
    tf.keras.layers.Reshape((30, 30, 32)),
    tf.keras.layers.MaxPooling2D((2, 2)),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Reshape((128, 225)),
    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(96, return_sequences=True, return_state=False)),
    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(96)),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(192, activation='relu'),
    tf.keras.layers.Dense(10),
])

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Optimizer
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
optimizer = tf.keras.optimizers.Nadam(
    learning_rate=0.0001, beta_1=0.9, beta_2=0.999, epsilon=1e-07,
    name='Nadam'
) # 0.00001

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Loss Fn
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""                               
lossfn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False)

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Model Summary
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
model.compile(optimizer=optimizer, loss=lossfn, metrics=['accuracy'])

"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
: Training
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""
history = model.fit(train_image_gen(), validation_data=train_image_gen(), batch_size=100, epochs=50 )

input( '..;.' )

输出：使用 ImageGenerator 进行训练，请以监控资源使用情况为目标。

Found 16 images belonging to 2 classes.
Epoch 1/50
2022-11-26 23:00:06.112861: I tensorflow/stream_executor/cuda/cuda_dnn.cc:368] Loaded cuDNN version 8100
16/16 [==============================] - 9s 146ms/step - loss: 1.1202 - accuracy: 0.4375 - val_loss: 0.7060 - val_accuracy: 0.5000
Epoch 2/50
16/16 [==============================] - 1s 57ms/step - loss: 0.7892 - accuracy: 0.3125 - val_loss: 0.6961 - val_accuracy: 0.5000
Epoch 3/50
 3/16 [====>.........................] - ETA: 0s - loss: 0.6903 - accuracy: 0.6667T

Answer 2

你必须使用 yield 而不是 return

def create_training_data():
    for category in CATEGORIES:  

        path = os.path.join(train_dir,category)  
        class_num = CATEGORIES.index(category)  

        for img in tqdm(os.listdir(path)):  
            try:
                img_train = cv2.imread(os.path.join(path,img))
                img_mean = cv2.blur(reduced_img_train,(9,9))
                yield [img_mean, class_num]  
            except Exception as e:
                pass

dataset = tf.data.Dataset.from_generator(create_training_data, output_types=(tf.float32 , tf.int32))