Exploiting Image Resolution Holistically in Computer Vision Models