초록
Over the past few decades, geospatial objects have been extensively recognized as significant components in remote sensing applications, including environmental monitoring, urban planning, and defense. Particularly, accurate segmentation of objects has aimed at meaningful observations from aerial imagery, leading to the necessity of deep learning-based methodologies. However, conventional deep learning-based segmentation methodologies exhibit limited generalization capabilities across diverse geographical domains due to inherent variations in regional characteristics and data distribution shifts. Furthermore, most existing approaches strongly rely on static, pre-trained models lacking the adaptability to handle previously unseen data. To alleviate these limitations, we propose a novel Few-shot Semi-Online Adaptation framework incorporating interactive user feedback to iteratively refine segmentation outputs. By leveraging online learning and test-time adaptation, our approach enables models to continuously be accurate based on minimal user corrections, ensuring flexibility and adaptability to new environments. Experimental results demonstrate that our method effectively enhances the segmentation accuracy with minimal user intervention, bridging the gap between automated segmentation and domain-specific expertise. Our research contributes to the development of interactive, user-adaptive segmentation models to facilitate geospatial analysis more efficiently and reliably.
김영수
학부연구생
이경수
조교수
연합학습 및 표현 학습을 사용한 medical 도메인에서의 딥 러닝 기법, 도메인 적응 및 테스트 타임 학습과 같은 딥 러닝 기반 컴퓨터 비전 응용, 이미지 처리 및 이미지-텍스트 캡셔닝을 포함한 의료 응용의 딥 러닝 기반 진단 등을 포함합니다.