What is the behaviour of keras preprocessing layer for data augmentation?

Keras
, ,
1

Say I have following preprocessing layers:

model = tf.keras.Sequential([
  # Add the preprocessing layers you created earlier.
  data_augmentation, # this will perform only rotation
  layers.Conv2D(16, 3, padding='same', activation='relu'),
  layers.MaxPooling2D(),
  # Rest of your model.
])

For example I have 30 images, then my model will be trained with 30 preprocessed (here rotation) images or it will be trained with 60 images (30 original images + 30 rotated (augmented) images)?.

If i want to increase the size of the dataset and former is the case then I need to use preprocessing layers seperately?

2

I’d assume that you now have n_images * n_rotations since the rotation at each angle will likely be random i.e you also have the original.