The study introduces a new set of multimodal remote sensing benchmark datasets (C2Seg dataset) for cross-city semantic segmentation and HighDAN, a high-resolution domain adaptation network. The AI models often face performance bottlenecks when dealing with cross-city or regional studies due to the lack of diverse remote sensing information and advanced solutions with high generalization ability. The C2Seg dataset includes two cross-city scenes, Berlin-Augsburg (Germany) and Beijing-Wuhan (China). HighDAN is designed to enhance the AI model’s generalization ability from multi-city environments. The study’s experiments show HighDAN’s superior segmentation performance and generalization ability compared to other competitors.
Publication date: 29 Sep 2023
Project Page: https://github.com/danfenghong
Paper: https://arxiv.org/pdf/2309.16499
