Caffe preprosessing training data

Question

I created a dataset of images and xml label files, where images are original size and alot higher resolution than what the network will handle. (2700x900 where as the net handle