Do we need to preprocess, i.e. resize images to a fixed predefined image input resolution (for example 256*256) on our own, in order to train properly?

[hedaniel7]

Hello there!

Let's say you have samples of picture of cones on the street in varying resolutions. If you would automatically resize, rescale them using a default cropping and padding algorithm to fit a predefined resolution, this would in some case distort the original content. However, if we would preprocess our images on our own in such a way to avoid that, would that result in a better object detection?

Or is the default resizing algorithm already good enough? In other words: Do we need to preprocess on our own for a better result?

Thank you in advance.

Daniel