22. Image Augmentation in Keras

If you have not yet launched a GPU-enabled server with AWS, you are strongly encouraged to do so before running the notebook from this video. While it is possible to train the notebook on your CPU, an AWS GPU instance will be much faster.

The Jupyter notebook described in the video can be accessed from the aind2-cnn GitHub repository. Navigate to the cifar10-augmentation/ folder and open cifar10_augmentation.ipynb.

Note on steps_per_epoch

Recall that fit_generator took many parameters, including

steps_per_epoch = x_train.shape[0] / batch_size

where x_train.shape[0] corresponds to the number of unique samples in the training dataset x_train. By setting steps_per_epoch to this value, we ensure that the model sees x_train.shape[0] augmented images in each epoch.

Optional Resources

• Read this great blog post that visualizes augmentations of the MNIST dataset.
• Check out this detailed implementation that uses augmentation to boost performance on a Kaggle dataset.
• Read the Keras documentation on the ImageDataGenerator class.