Constraining pseudo‐label in self‐training unsupervised domain adaptation with energy‐based model

Deep learning is usually data starved, and the unsupervised domain adaptation (UDA) developed to introduce knowledge in labeled source unlabeled target domain. Recently, deep self-training presents a powerful means for UDA, involving an iterative process of predicting then taking confident predictions as hard pseudo-labels retraining. However, are unreliable, thus easily leading deviated solutions with propagated errors. In this paper, we resort energy-based model constrain training sample energy function minimization objective. It can be achieved via simple additional regularization or loss. This framework allows us gain benefits model, while retaining strong discriminative performance following plug-and-play fashion. The convergence property its connection classification expectation investigated. We deliver extensive experiments on most popular large-scale UDA benchmarks image well semantic segmentation demonstrate generality effectiveness.