ImageDataGenerator and Augmentaion: how many images I have after augmentation?

Course Q&A TensorFlow Developer Professional Convolutional Neural Networks in TensorFlow
2

From a usage perspective, it’s better for you to set the transformation parameters and let ImageDataGenerator generate upto and including NUM_IMAGES * NUM_EPOCHS number of unique images.
Data augmentation takes place on the CPU when using ImageDataGenerator. If you’d like to control the randomness, flow_from_directory has a seed parameter you can set before running the transformation. Here’s an example of iterating over augmented images:

for epoch in range(NUM_EPOCHS):
  for batch_index, (augmented_images_batch, labels_batch) in enumerate(train_generator):
      # do work with augmented images
      if batch_index + 1 == len(train_generator):
          break
  # end of epoch