We welcome the ascension of our artificial overlords!

The new builds upon the old, eventually becoming unrecognizable to each other. By then, they are merely linked by their shared history.

While OpenCV and Keras offer robust data augmentation capabilities, the ever-evolving landscape of deep learning has spawned a plethora of specialized libraries designed to enhance your image and data processing workflows.

Why perform data augmentation?

Referring the past few chapters about OpenCV, data augmentation involves applying various transformations to existing images to create new, synthetic training samples. Doing so benefits the model in the context of:

• Increased Dataset Size: significantly increasing the size of the training dataset, improving model performance and generalization.

• Improved Robustness: helping the model become more robust to variations in real-world data, such as changes in lighting, pose, or occlusion.

• Regularization: acting as a form of regularization, preventing the model from overfitting to the training data.

Alternatives to data augmentation

Besides the popular OpenCV and Keras' staple ImageDataGenerator, there are other libraries and frameworks that offer additional options or specialized functionalities:

• imgaug: allows users to define their own augmentation techniques or combine multiple techniques. Often considered more advanced and efficient than OpenCV. A good choice for users who need a high degree of customization and flexibility.

• Albumentations: offers a wide range of augmentation techniques ( i.e., geometric transformations, color transformations, cropping). Supports both image and video augmentation. Ideal for those seeking a wide range of augmentation techniques and efficiency.

Ultimately, the best library for your project will depend on your specific needs, the complexity of your augmentation pipeline, and the deep learning framework you are using. If you think one augmenter trumps all others in every aspect, explain why the others are still being used as of today.