I wonder how the original data is separated into pieces of 384 * 384 * 1. Can I piece them back? My idea is to revise the __getitem__ in dataloader, such that it will randomly sample a window from the big images. Many thanks!