Currently it seems like you can only specific the size of the height and width dimensions together for a square shape. I would like to be able to independently specify different height and width values, for certain kinds of training data that for example uses a 1:2 ratio (768x1536) and requires that the entire image be visible to the model during each training step.
Currently it seems like you can only specific the size of the height and width dimensions together for a square shape. I would like to be able to independently specify different height and width values, for certain kinds of training data that for example uses a 1:2 ratio (768x1536) and requires that the entire image be visible to the model during each training step.