EUVIMEDEUVIMED entdecken
Crossref

Scribble-Guided Hierarchical Prompt for SAM-Based Weakly Supervised Salient Object Detection

Fen Xiao, Ruozhuo Huang, Zhenwei Wu, Xieping Gao

IEEE Transactions on Circuits and Systems for Video Technology · 2026-08 · 36 (8) · 11761-11772

DOI: 10.1109/tcsvt.2026.3682902 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-08. Verfasst von Fen Xiao und 3 weiteren Autor:innen. Die Arbeit nennt 65 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Scientific Research Fund of Hunan Provincial Education Department, Postgraduate Scientific Research Innovation Project of Hunan Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Weakly supervised salient object detection (WSOD) employs sparse annotations to encode localized saliency cues, which aligns well with the prompt-driven paradigm of the Segment Anything Model (SAM). However, existing SAM-based WSOD approaches often suffer from segmentation artifacts over- or under-segmentation, due to their use of one-shot unrefined prompts derived from sparse annotations. To address this, we propose a scribble-guided hierarchical prompt framework (SHP-SAM) that progressively builds and optimizes hierarchical prompts including points, boxes and masks through scribble annotations and segmentation feedback to guide SAM for effective WSOD. Specifically, a Scribble Decomposition Module (SDM) is introduced to decompose scribbles into coherent subunits via saliency-guided clustering, preserving strong responses while revealing weakly activated salient regions. For each sub-scribble, the Hierarchical Prompt Generation Module (HPGM) constructs hierarchical prompts through initial generation and iterative refinement to guide SAM in segmenting local salient regions, and the resulting masks are aggregated into a global pseudo-label. Finally, we introduce a Boundary Regression Module (BRM) to enhance boundary localization by incorporating an edge supervision branch and filtering out edge cues inconsistent with salient regions. Extensive experiments on five benchmarks demonstrate that SHP-SAM outperforms existing WSOD methods and achieves performance comparable to fully supervised approaches. Our code and results are available at https://github.com/XTU-ICIPL/SHP-SAM

Förderung

  • National Natural Science Foundation of China
  • Scientific Research Fund of Hunan Provincial Education Department
  • Postgraduate Scientific Research Innovation Project of Hunan Province

Zitieren

APA

Xiao, F., Huang, R., Wu, Z., & Gao, X. (2026). Scribble-Guided Hierarchical Prompt for SAM-Based Weakly Supervised Salient Object Detection. IEEE Transactions on Circuits and Systems for Video Technology, 36(8), 11761-11772. https://doi.org/10.1109/tcsvt.2026.3682902

Vancouver

Xiao F, Huang R, Wu Z, Gao X. Scribble-Guided Hierarchical Prompt for SAM-Based Weakly Supervised Salient Object Detection. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(8):11761-11772. doi:10.1109/tcsvt.2026.3682902

BibTeX
@article{Xiao2026,
  author = {Xiao, Fen and Huang, Ruozhuo and Wu, Zhenwei and Gao, Xieping},
  title = {Scribble-Guided Hierarchical Prompt for SAM-Based Weakly Supervised Salient Object Detection},
  journal = {IEEE Transactions on Circuits and Systems for Video Technology},
  year = {2026},
  volume = {36},
  number = {8},
  pages = {11761--11772},
  doi = {10.1109/tcsvt.2026.3682902}
}
Kostenlos starten →