0
$\begingroup$

For the purposes of training a Convolutional Neural Network Classifier, should image augmentation be done before or after resizing the training images?

To reduce file size and speed up training time, developers often resize training images to a set height and width using something like PIL (Python Imaging Library).

If the images are augmented (to increase training set size), should it be done before or after resizing the members of the set?

For simplicity sake, it would probably be faster to augment the images after resizing, but I am wondering if any useful data is lost in this process. I assume it may depend on the method used to resize the images (cropping, scaling technique, etc.)

$\endgroup$
1

1 Answer 1

0
$\begingroup$

As it is told in PIL documentation

It uses some filters to resize images.And those filters are explained here uses mostly numerical methods as I see. So it is approximating the image data input. Which means you are right about data loss. But here might be the question would it change the data so much If it is done after augmentation or before?

Since in numerical methodic approaches the more values means more valid approximation in general.

So it might be beneficial to augment and then resize.

But one should look for its real answer. Mine is just a thought experiment. You will get better proven answers

$\endgroup$
1
  • $\begingroup$ Awesome, thanks! I'm going to look around for similar studies done, and perhaps perform my own. $\endgroup$ Sep 7, 2020 at 20:36

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .