Learn to perform data augmentation with the AutoAugment technique.

We'll cover the following

The AutoAugment is another standalone data augmentation technique that works similarly to RandAugment.

AutoAugment class complements other data augmentation techniques. Most of the top-performing models use AutoAugment in their training pipelines.

The AutoAugment augmentation

The AutoAugment augmentation accepts a PIL.Image as input, which we can configure with the config_str variable.

However, the image transformations are bundled as policies and sub-policies instead. The original policy consists of 25 sub-policies. The AutoAugment module will iterate through each sub-policy and perform the corresponding image transformations.

Like before, we need to load the image with PIL before data augmentation. Next, call the auto_augment_transform function from module:

Get hands-on with 1200+ tech skills courses.