Abstract: Multi-task learning constitutes the prevalent paradigm in numerous vision applications that cast an eye on runtime efficiency. At present however, deep multi-task networks are limited in ...