In the field of domain adaptation, a trade-off exists between the model performance and the number of target domain annotations. Active learning, maximizing model performance with few informative labeled data, comes in handy for such a scenario. In this work, we present ADeADA, a general active domain adaptation framework for semantic segmentation. To adapt the model to the target domain with minimum queried labels, we propose acquiring labels of the samples with high probability density in the target domain yet with low probability density in the source domain, complementary to the existing source domain labeled data. To further facilitate the label efficiency, we design an adaptive budget allocation policy, which dynamically balances the labeling budgets among different categories as well as between density-aware and uncertainty-based methods. Extensive experiments show that our method outperforms existing active learning and domain adaptation baselines on two benchmarks, GTA5 -> Cityscapes and SYNTHIA -> Cityscapes. With less than 5% target domain annotations, our method reaches comparable results with that of full supervision.

D2ADA is a framework for active domain adaptation in semantic segmentation, proposing an approach to acquire labels of samples with high probability density in the target domain yet with low probability density in the source domain, and designing a dynamic scheduling policy to adjust labeling budgets between domain exploration and model uncertainty that outperforms existing baselines with less than 5% target domain annotations.

D2ADA: 动态密度感知的主动域自适应语义分割