Visual learning in-the-wild with limited supervision: the role of disentangled factors, auxiliary domains, and active supervision