Hi, is there a way to train using a dataset of images with shape that is not a power of 2? Such as 400x400 or even 420x400? Thank you Davide