【tensorflow2.0 图片数据】tensorflow中准备图片数据的常用方案

在tensorflow中准备图片数据的常用方案有两种，第一种是使用tf.keras中的ImageDataGenerator工具构建图片数据生成器。

第二种是使用tf.data.Dataset搭配tf.image中的一些图片处理方法构建数据管道。

第一种方法更为简单

from keras.preprocessing.image import ImageDataGenerator

train_dir = 'cifar2_datasets/train'
test_dir = 'cifar2_datasets/test'

# 对训练集数据设置数据增强
train_datagen = ImageDataGenerator(
            rescale = 1./255,
            rotation_range=40,
            width_shift_range=0.2,
            height_shift_range=0.2,
            shear_range=0.2,
            zoom_range=0.2,
            horizontal_flip=True,
            fill_mode='nearest')

# 对测试集数据无需使用数据增强
test_datagen = ImageDataGenerator(rescale=1./255)


train_generator = train_datagen.flow_from_directory(
                    train_dir,
                    target_size=(32, 32),
                    batch_size=32,
                    shuffle = True,
                    class_mode='binary')

test_generator = test_datagen.flow_from_directory(
                    test_dir,
                    target_size=(32, 32),
                    batch_size=32,
                    shuffle = False,
                    class_mode='binary')


from keras import models,layers,optimizers
from  keras import backend as K

K.clear_session()
model = models.Sequential()
model.add(layers.Flatten(input_shape = (32,32,3)))
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

model.compile(loss='binary_crossentropy',
            optimizer=optimizers.RMSprop(lr=1e-4),
            metrics=['acc'])


# 计算每轮次需要的步数 
import numpy as np 
train_steps_per_epoch  = np.ceil(10000/32)
test_steps_per_epoch  = np.ceil(2000/32)

# 使用内存友好的fit_generator方法进行训练
history = model.fit_generator(
        train_generator,
        steps_per_epoch = train_steps_per_epoch,
        epochs = 5,
        validation_data= test_generator,
        validation_steps=test_steps_per_epoch,
        workers=1, # 读取数据的进程数
        use_multiprocessing=False #linux上可使用多进程读取数据
        )

第二种方式，TensorFlow的原生方法，更加灵活，使用得当的话也可以获得更好的性能。：

import tensorflow as tf
from tensorflow.keras import datasets,layers,models
import datetime
import pandas as pd
import matplotlib.pyplot as plt

BATCH_SIZE = 100

# 自定义导入图片函数
def load_image(img_path,size = (32,32)):
    label = tf.constant(1,tf.int8) if tf.strings.regex_full_match(img_path,".*/automobile/.*") else tf.constant(0,tf.int8)
    img = tf.io.read_file(img_path)
    img = tf.image.decode_jpeg(img) #注意此处为jpeg格式
    img = tf.image.resize(img,size)/255.0
    return(img,label)




#使用并行化预处理num_parallel_calls 和预存数据prefetch来提升性能
ds_train = tf.data.Dataset.list_files("./data/cifar2/train/*/*.jpg") \
           .map(load_image, num_parallel_calls=tf.data.experimental.AUTOTUNE) \
           .shuffle(buffer_size = 1000).batch(BATCH_SIZE) \
           .prefetch(tf.data.experimental.AUTOTUNE)

ds_test = tf.data.Dataset.list_files("./data/cifar2/test/*/*.jpg") \
           .map(load_image, num_parallel_calls=tf.data.experimental.AUTOTUNE) \
           .batch(BATCH_SIZE) \
           .prefetch(tf.data.experimental.AUTOTUNE)

#清空会话
tf.keras.backend.clear_session()

# 定义模型结构
inputs = layers.Input(shape=(32,32,3))
x = layers.Conv2D(32,kernel_size=(3,3))(inputs)
x = layers.MaxPool2D()(x)
x = layers.Conv2D(64,kernel_size=(5,5))(x)
x = layers.MaxPool2D()(x)
x = layers.Dropout(rate=0.1)(x)
x = layers.Flatten()(x)
x = layers.Dense(32,activation='relu')(x)
outputs = layers.Dense(1,activation = 'sigmoid')(x)
model = models.Model(inputs = inputs,outputs = outputs)
print(model.summary())


# 定义日志目录
logdir = "./data/keras_model/" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
tensorboard_callback = tf.keras.callbacks.TensorBoard(logdir, histogram_freq=1)

# 模型编译
model.compile(
        optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
        loss=tf.keras.losses.binary_crossentropy,
        metrics=["accuracy"]
    )

# 模型训练
history = model.fit(ds_train,epochs= 10,validation_data=ds_test,
                    callbacks = [tensorboard_callback],workers = 4)



dfhistory = pd.DataFrame(history.history)
dfhistory.index = range(1,len(dfhistory) + 1)
dfhistory.index.name = 'epoch'



# 可视化
def plot_metric(history, metric):
    train_metrics = history.history[metric]
    val_metrics = history.history['val_'+metric]
    epochs = range(1, len(train_metrics) + 1)
    plt.plot(epochs, train_metrics, 'bo--')
    plt.plot(epochs, val_metrics, 'ro-')
    plt.title('Training and validation '+ metric)
    plt.xlabel("Epochs")
    plt.ylabel(metric)
    plt.legend(["train_"+metric, 'val_'+metric])
    plt.show()


plot_metric(history,"loss")
val_loss,val_accuracy = model.evaluate(ds_test,workers=4)
print(val_loss,val_accuracy)