Custom BatchIterator

[ProgRoman]

Hello! When using this code

from nolearn.lasagne import BatchIterator
class SimpleBatchIterator(BatchIterator):
    def transform(self, Xb, yb):
        # The 'incomming' and outcomming shape is (10, 1, 28, 28)
        Xb, yb = super(SimpleBatchIterator, self).transform(Xb, yb)
        return manipulateTrainingData(Xb), yb #<--- Here we do the manipulations of the training set
# Setting the new batch iterator
net1Aug = createNet()
net1Aug.max_epochs = epochs_augmentation
net1Aug.batch_iterator_train = SimpleBatchIterator(256)

The original image is completely replaced by the new image after a call to the method
transform(self, Xb, yb)
ie training sets extends (adding new modified image)
or training set remains the same size replaced only the images themselves?