EUVIMEDEUVIMED entdecken
Crossref

A Large-Scale Benchmark Dataset and Reconstruction-Guided Light Field Salient Object Detection

Linxuan He, Chao Wang, Pingping Zhang, Zeyu Wang, Pengjie Wang

IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 17748-17760

DOI: 10.1109/tcsvt.2026.3718658 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Verfasst von Linxuan He und 4 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Gefördert von: Research Fund of the Liaoning Provincial Education Department, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Salient Object Detection (SOD) aims to simulate the human visual attention mechanism to automatically locate and segment the most visually attractive objects in complex scenes. However, existing light field SOD datasets suffer from limited data quality and insufficient category diversity, which hinders the development of light field SOD. Moreover, current methods still face challenges in fully exploiting the valuable depth cues and diverse focus features within focal stacks. To address these problems, we construct a large-scale benchmark dataset for light field SOD, named DLMU-LF. It contains 5,037 samples with comprehensive annotations and covers diverse object categories. Based on this dataset, we propose a novel framework, Reconstruction-guided Light Field Salient Object Detection Network (RLFSNet), which leverages focal stack reconstruction of all-in-focus images to assist salient object detection. Specifically, RLFSNet introduces an all-in-focus reconstruction branch to synthesize complete scene representations and alleviate the discreteness of focal stacks. In addition, we design a Hierarchical Selective Fusion (HSF) module and a Frequency Guided Feature Fusion (FGFF) module to model cross-slice dependencies and enhance cross-modal features. Extensive experiments demonstrate that RLFSNet achieves superior performance over state-of-the-art methods across multiple benchmark datasets, while maintaining stable performance and strong generalization in complex scenarios. The DLMU-LF dataset and source code are publicly available at https://github.com/VMS203/DLMU-LF.

Förderung

  • Research Fund of the Liaoning Provincial Education Department
  • National Natural Science Foundation of China

Zitieren

APA

He, L., Wang, C., Zhang, P., Wang, Z., & Wang, P. (2026). A Large-Scale Benchmark Dataset and Reconstruction-Guided Light Field Salient Object Detection. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 17748-17760. https://doi.org/10.1109/tcsvt.2026.3718658

Vancouver

He L, Wang C, Zhang P, Wang Z, Wang P. A Large-Scale Benchmark Dataset and Reconstruction-Guided Light Field Salient Object Detection. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):17748-17760. doi:10.1109/tcsvt.2026.3718658

BibTeX
@article{He2026,
  author = {He, Linxuan and Wang, Chao and Zhang, Pingping and Wang, Zeyu and Wang, Pengjie},
  title = {A Large-Scale Benchmark Dataset and Reconstruction-Guided Light Field Salient Object Detection},
  journal = {IEEE Transactions on Circuits and Systems for Video Technology},
  year = {2026},
  volume = {36},
  number = {11},
  pages = {17748--17760},
  doi = {10.1109/tcsvt.2026.3718658}
}
Kostenlos starten →