So when performing image/data augmentation, is it possible to put both the original image and the augmented image into the training set?
For example, when I augment a cat image, the augmented image is the only one being trained by the model. But wouldn’t it be better to put both image into the system? After all, they should be considered two separate pieces of data.
They are both present in the training dataset.
As a little addittion to @gent.spah’s response, when you perform any kind of data augmentation (image in this case), you define a series of transformations that are applied with a certain probability. That means they will be applied just in some batches. So both images, augmented and original, will be used for training.
it is using both in-memory