Domain Adaptive Semantic Segmentation through Photorealistic Enhancement of Video Game
Kaito Nakajima, Takafumi Katayama, Tian Song, Xiantao Jiang, Takashi Shimamoto · 2022
Unsupervised domain adaptation is considered as an effective technique to reduce the large amount supervised data. In order to solve this problem, unsupervised domain adaptation is considered to be an effective technique. In this work, three types of domain adaptation: image-level domain adaptation, inter-domain adaptation, and intra-domain adaptation are introduced to achieve better semantic segmentation accuracy. The proposed method achieved an mean IoU of 45.0%. Furthermore, by combining the proposed method with intra-domain adaptation, an mean IoU improvement of 1.2% is achieved compared to previous work.