this amazing performance improvments kill this area!!!!!!!!!!

[jingzhengli]

this amazing performance improvements kill this area!!!!!!!!!!

[panzhang0212]

I think there is still a large gap with full supervised learning.

1. recent self-supervised learning could get comparable results with supervised learning, why UDA is still worse than full supervised learning.
2. ProDA follows self-training. Another solution for UDA is gap minimization, which is more general and needs more works to improve.