EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Semi-Supervised 3-D Point Cloud Semantic Segmentation Using Multiscale Graph Attention and Class-Incremental Learning

Hui Ye, Wei Guo, Liang Ma, Xiangyi Shi, Jianqiang Ma

IEEE Access · 2026 · 14 · 45423-45436

DOI: 10.1109/access.2026.3674775 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Hui Ye und 4 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Class-incremental learning for 3D point cloud segmentation requires dense annotations for novel classes, which is prohibitively expensive and impractical for real-world applications with scarce labels. To bridge this gap, we propose a Semi-Supervised Class-Incremental Learning (SSCIL) framework that significantly reduces annotation costs. Our framework introduces three key innovations: 1) relaxed geometry-aware feature-relation distillation module that transfers knowledge with relaxed constraints; 2) unsupervised learning module, comprising color reconstruction and a view consistency module (CR-VC), which leverages unlabeled data to learn robust and generalizable features; and 3) multi-scale dynamic graph attention (MDG) backbone that captures both local geometric details and global contextual relationships. Extensive experiments on S3DIS and ScanNet benchmarks demonstrate that our method achieves competitive performance compared to the fully-supervised 3DPC-CISS, attaining approximately 97% of its mIoU on the $C_{\mathrm {novel}}=5$ , $S_{0}$ split while using only 10% of the labeled novel-class data. Our work establishes a practical and efficient paradigm for label-efficient 3D scene understanding, with a $10\times $ reduction in annotation effort, making incremental learning viable in label-scarce scenarios. Our code will be made publicly available at https://github.com/maliang07/SSCIL2 upon publication.

Zitieren

APA

Ye, H., Guo, W., Ma, L., Shi, X., & Ma, J. (2026). Semi-Supervised 3-D Point Cloud Semantic Segmentation Using Multiscale Graph Attention and Class-Incremental Learning. IEEE Access, 14, 45423-45436. https://doi.org/10.1109/access.2026.3674775

Vancouver

Ye H, Guo W, Ma L, Shi X, Ma J. Semi-Supervised 3-D Point Cloud Semantic Segmentation Using Multiscale Graph Attention and Class-Incremental Learning. IEEE Access. 2026;14:45423-45436. doi:10.1109/access.2026.3674775

BibTeX
@article{Ye2026,
  author = {Ye, Hui and Guo, Wei and Ma, Liang and Shi, Xiangyi and Ma, Jianqiang},
  title = {Semi-Supervised 3-D Point Cloud Semantic Segmentation Using Multiscale Graph Attention and Class-Incremental Learning},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {45423--45436},
  doi = {10.1109/access.2026.3674775}
}
Kostenlos starten →