Hands-On Machine Learning with C++

Data augmentation

The data augmentation process can be treated as regularization because it adds some prior knowledge about the problem to the model. This approach is common in computer vision tasks such as image classification or object detection. In such cases, when we can see that the model begins to overfit and does not have enough training data, we can augment the images we already have to increase the size of our dataset and provide more distinct training samples. Image augmentations are random image rotations, cropping and translations, mirroring flips, scaling, and proportion changes.