C1W4 Lab 1 - resizing images

[fdam]

In the lab, data generator read images and resize them:

target_size=(300, 300), # All images will be resized to 300x300

A the same way, at the prediction time, a resizing is applied:

img = load_img(path, target_size=(300, 300))

But can this kind of image preprocessing introduce some sort of distortion into the image causing poor performance in both training and prediction?

[gent.spah]

Yes it can!

But hopefully is minimal.

Any kind of preprocessing alters the original features but hopefully is not a big alteration to deviate from what the model should learn.