Generators

In the previous chapter, we discussed how big datasets could lead to problems in training due to the limitations in RAM. This problem is a bigger issue when working with images. Keras has implemented generators that help us get batches of input images and their corresponding labels while training on the fly. These generators also help us perform data augmentation on images before using them for training. First, we will see how we can make use of the ImageDataGenerator class to generate augmented images for our model.

To implement data augmentation, we just need to change our Exercise 3 code a little bit. We will substitute model.fit() with the following:

BATCH_SIZE = 32

aug = ImageDataGenerator(rotation_range=20,

width_shift_range=0.2, height_shift_range=0.2,

shear_range=0.15, zoom_range=0.15,

horizontal_flip=True, vertical_flip=True,

fill_mode=”nearest”)

 

log = model.fit_generator(

aug.flow(x_train, y_train, batch_size= BATCH_SIZE),

validation_data=( x_test, y_test...