Unsupervised Domain Adaptation for Semantic Segmentation by Content Transfer

In this paper, we tackle the unsupervised domain adaptation (UDA) for semantic segmentation, which aims to segment unlabeled real data using labeled synthetic data. The main problem of UDA segmentation relies on reducing gap between image and image. To solve problem, focused separating information in an into content style. Here, only has cues style makes gap. Thus, precise separation leads effect as supervision even when learning with make best effect, propose a zero-style loss. Even though perfectly extract domain, another challenge, class imbalance still exists segmentation. We address by transferring contents tail classes from domain. Experimental results show that proposed method achieves state-of-the-art performance major two settings.