How well does a convolution network works with non-square resolution?

Hello everyone,

For my project I’m trying to match high resolution images to their defined labels. You can find the dataset here: https://zenodo.org/record/3477553/#.XvUUmWhKiUk

The current problem that I’m facing is that the images are in 4k resolution (3840 x 2160) which makes them rectangular. I have two questions. How well does a convolution network works with non-square resolution? I know that I need to some experimenting myself, but what could happen if I downscale the image to a square resolution like 128x128. Any help on this issues would be greatly appreciate it. Thank you.

1 Like