Confusion in Keras data augmentation process

Question

I tried this experiment:

1) I have two classes 'moon' and 'star', and each have 3 images.

2) I tried creating data augmented images as described below.

trainingDataGenerator = ImageDataGenerator(rotation_range=10)

trainGenerator = trainingDataGenerator.flow_from_directory(
                        trainingDataDir,
                        target_size=(28,28),
                        batch_size = 1,
                        seed=7,
                        class_mode='binary',
                        save_to_dir='data/augmented')

# Define Model Here. 

# fit_generator is like this:
   model.fit_generator(
                  trainGenerator,
                  steps_per_epoch = 5,
                  epochs=1)

When I looked at the data/augmented directory, the number of images generated is 11 (eleven).

My Question:

Total Number of images per epoch is = steps_per_epoch * batch_size

This means that in my case it should generate 5*1 = 5 images for 1 epoch. But I see 11 images in the augmented directory.

Could someone kindly explain how this data augmented images are counted?

Do u have any sample code which works directly with flow_from_directory rather cifar. Most of the examples are pertaining to cifar only. — DrunkenMaster, Jul 15 '17 at 09:59
https://stackoverflow.com/questions/36986815/what-is-the-parameter-max-q-size-used-for-in-model-fit-generator This answer will be helpful. — raul007007, Oct 26 '18 at 02:57

Confusion in Keras data augmentation process

0 Answers0