tensoflow: 从文件夹加载数据管道的两种方式

 1. ImageDataGenerator

train_datagen = ImageDataGenerator(rescale=1./255, 
                                   zoom_range=0.20, 
                                   fill_mode="nearest")
print(type(train_datagen)) # >>> <class 'keras.preprocessing.image.ImageDataGenerator'>

validation_datagen = ImageDataGenerator(rescale=1./255)
      
train_generator = train_datagen.flow_from_directory(train_path, 
                                                    target_size=(img_rows, img_cols), 
                                                    batch_size=batch_size, 
                                                    class_mode='categorical', 
                                                    subset='training')
print(type(train_generator)) # >>> <class 'keras_preprocessing.image.directory_iterator.DirectoryIterator'>
validation_generator = validation_datagen.flow_from_directory(validation_path, 
                                                              target_size=(img_rows, img_cols), 
                                                              batch_size=batch_size, 
                                                              class_mode=None,  # only data, no labels 
                                                              shuffle=False)   
 
history = model.fit_generator(train_generator, 
                              steps_per_epoch=len(train_generator), 
                              epochs=epochs)
    
predictions = model.predict_generator(validation_generator,
                                          steps=len(validation_generator),
                                          verbose=1)

2.  image_dataset_from_directory

train_dataset = image_dataset_from_directory(train_dir,
                                             shuffle=True, # 默认: True
                                             batch_size=BATCH_SIZE, # 默认:（256 x 256）
                                             image_size=IMG_SIZE) # 默认: 32

validation_dataset = image_dataset_from_directory(validation_dir,
                                                  label_mode='binary',
                                                  shuffle=True, 
                                                  batch_size=BATCH_SIZE, 
                                                  image_size=IMG_SIZE)
print(type(validation_dataset)) # >>> <class 'tensorflow.python.data.ops.dataset_ops.BatchDataset'>
test_dataset = image_dataset_from_directory(test_dir, 
                                            label_mode='binary',
                                            shuffle=True, 
                                            image_size=IMG_SIZE)
# print(dir(test_dataset)) # dir(object)：返回对象的属性和方法。
print(test_dataset.class_names) # 数据集属性中类的名称

history = model.fit(train_dataset,
                    epochs=initial_epochs, 
                    validation_data=validation_dataset)