Kind of a silly question, but I read that data Augmentation can be used in order to solve problem of small datasets. In my case, I've got a dataset with 5 different classes and around 2k examples per class. After performing various image transformations, do I merge the augmented dataset with original dataset into a new, bigger (around 20k image) dataset, or did I misunderstood this procedure?
$\begingroup$
$\endgroup$
asked May 6, 2020 at 21:07
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Yes, use all the data you can.
The goal behind data augmentation is to indeed create more data points from a smaller set of instances. Therefore, you definitely want to keep your original data in there, since especially, this is the only "pure" data you have.
answered May 7, 2020 at 7:32