본문 바로가기
  • Home

Open-World Object Detection and Segmentation: Current Challenges and Emerging Opportunities

  • Journal of Internet of Things and Convergence
  • Abbr : JKIOTS
  • 2025, 11(6), 27
  • Publisher : The Korea Internet of Things Society
  • Research Area : Engineering > Computer Science > Internet Information Processing
  • Received : October 12, 2025
  • Accepted : December 15, 2025
  • Published : December 31, 2025

Muhammad Iqubal Ali 1 In-Chul Han 2 KIM, SOO KYUN 2

1제주대학교 컴퓨터공학과
2제주대학교

Accredited

ABSTRACT

Real-world vision systems often encounter categories unseen during training, challenging the closed-set limitation of traditional detectors. This review summarizes advances in Open-World Object Detection (OWOD) and its extension to open-world segmentation, where models must recognize known classes, reject novel instances, and incrementally learn new categories. We categorize OWOD approaches into pseudo-labeling, class-agnostic proposal learning, metric-based novelty scoring, and hybrid methods leveraging vision–language or video cues, connecting them to open-world semantic and instance segmentation. Key challenges include proposal bias, open-set misclassification, continual learning degradation, and inconsistent evaluation. Finally, we highlight future directions such as integration with 3D perception, self/semi-supervised learning, and calibrated uncertainty, outlining a roadmap toward reliable open-world perception.

Citation status

* References for papers published after 2024 are currently being built.