This research focuses on the creation of a new set of multimodal remote sensing benchmark datasets for cross-city semantic segmentation tasks. The dataset, called the C2Seg, includes two cross-city scenes, Berlin-Augsburg (Germany) and Beijing-Wuhan (China). The researchers also propose a high-resolution domain adaptation network, HighDAN, to improve the AI model’s generalization ability across city environments. HighDAN can maintain the spatial topology of the studied urban scene and close the gap derived from massive differences in remote sensing image representations between different cities. The researchers conducted experiments on the C2Seg dataset, showing that HighDAN outperforms its competitors in terms of segmentation performance and generalization ability.
Publication date: 29 Sep 2023
Project Page: https://github.com/danfenghong
Paper: https://arxiv.org/pdf/2309.16499
