Efficient Visual Region Recognition in the Open World Scenarios
Journal of Computer Science and Artificial Intelligence · 27.03.2025 · 2 (3) · 40-49
DOI: 10.54097/j9maj395 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Computer Science and Artificial Intelligence, online erschienen am 27.03.2025. Verfasst von Jing Wang und 1 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Open-World Object Detection (OWOD) aims to address the core challenges that traditional detection models cannot dynamically adapt to unknown objects and continuously learn new categories. Most existing methods rely on visual similarity to discover new objects, yet they still face issues such as semantic disconnection and catastrophic forgetting in complex cross-domain scenarios, such as medical imaging and autonomous robotics. This paper proposes a causality-driven open-world object detection framework. By decoupling the functional causal features and the appearance features of objects, it realizes the discovery of unknown objects and incremental learning based on physical property reasoning. The core contributions include: constructing a feature enhancement model, which significantly improves the quality and expressiveness of image features; establishing a similarity calculation mechanism. By accurately calculating the similarity between text and images, it effectively avoids the over-confidence phenomenon in model prediction, thus improving the accuracy and reliability of detection; adopting a multi-granularity visual stream to conduct multi-dimensional and refined feature processing on image features, fully exploring the multi-level information in images. To comprehensively evaluate the performance of the proposed method, we conduct in-depth research on existing open-world benchmarks and extensively validate it on public benchmark datasets. The experimental results clearly demonstrate that this method achieves absolute performance gains in the detection tasks of unknown categories, fully demonstrating its strong generalization ability. This research result provides valuable reference and inspiration for the further development of the open-world object detection field, and is expected to promote technological progress and innovation in this field.
Zitieren
APA
Wang, J., & Cao, Y. (2025). Efficient Visual Region Recognition in the Open World Scenarios. Journal of Computer Science and Artificial Intelligence, 2(3), 40-49. https://doi.org/10.54097/j9maj395
Vancouver
Wang J, Cao Y. Efficient Visual Region Recognition in the Open World Scenarios. Journal of Computer Science and Artificial Intelligence. 2025;2(3):40-49. doi:10.54097/j9maj395
BibTeX
@article{Wang2025,
author = {Wang, Jing and Cao, Yonghua},
title = {Efficient Visual Region Recognition in the Open World Scenarios},
journal = {Journal of Computer Science and Artificial Intelligence},
year = {2025},
volume = {2},
number = {3},
pages = {40--49},
doi = {10.54097/j9maj395}
}
EUVIMED