Can the jointremoval model train models of arbitrary resolution? What parameters need to be configured?