How to Train Computer Vision Models Efficiently