Data Augmentation Clarification

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

Hi Sir,

In the lecture video Data Augmentation, in the lecture slide naming, Implement Distortions using training slide we had below doubts like, can u please help to clarify.,

Doubt is mini batches of data will be used to pass for implementing training or only each distorted image used to pass for training ?

And one more doubt is thread concept will run implicitly or should we do programatically implement the thread concept ?

2

Hi Anbu,

In standard implementation, the original images are not used for training, but only augmented images are. Some of these augmented images will be quite similar to the original ones. See, for instance, this description.

Optimal threading through available CPUs and GPUs is a subject by itself. As an indication, you can have a look at this discussion.