EUVIMEDEUVIMED entdecken
Crossref

Reference Twice: A Simple and Unified Baseline for Few-Shot Instance Segmentation

Yue Han, Jiangning Zhang, Yabiao Wang, Chengjie Wang, Yong Liu, Lu Qi, Ming-Hsuan Yang, Xiangtai Li

IEEE Transactions on Pattern Analysis and Machine Intelligence · 2024-12 · 46 (12) · 9221-9238

DOI: 10.1109/tpami.2024.3421340 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024-12. Verfasst von Yue Han und 7 weiteren Autor:innen. Die Arbeit wurde bisher 18-mal zitiert (Crossref) und nennt 118 Literaturangaben. Gefördert von: Science and Technology Planning Project of Zhejiang Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Few-Shot Instance Segmentation (FSIS) requires detecting and segmenting novel classes with limited support examples. Existing methods based on Region Proposal Networks (RPNs) face two issues: 1) overfitting suppresses novel class objects and 2) dual-branch models require complex spatial correlation strategies to prevent spatial information loss when generating class prototypes. We introduce a unified framework, Reference Twice (RefT), to exploit the relationship between support and query features for FSIS and related tasks. Our three main contributions are: 1) a novel transformer-based baseline that avoids overfitting, offering a new direction for FSIS; 2) demonstrating that support object queries encode key factors after base training, allowing query features to be enhanced twice at both feature and query levels using simple cross-attention, thus avoiding complex spatial correlation interaction; and 3) introducing a class-enhanced base knowledge distillation loss to address the issue of DETR-like models struggling with incremental settings due to the input projection layer, enabling easy extension to incremental FSIS. Extensive experimental evaluations on the COCO dataset under three FSIS settings demonstrate that our method performs favorably against existing approaches across different shots, e.g., +8.2/+9.4 performance gain over state-of-the-art methods with 10/30-shots.

Förderung

  • Science and Technology Planning Project of Zhejiang Province

Zitieren

APA

Han, Y., Zhang, J., Wang, Y., Wang, C., Liu, Y., Qi, L., Yang, M. H., & Li, X. (2024). Reference Twice: A Simple and Unified Baseline for Few-Shot Instance Segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(12), 9221-9238. https://doi.org/10.1109/tpami.2024.3421340

Vancouver

Han Y, Zhang J, Wang Y, Wang C, Liu Y, Qi L, et al. Reference Twice: A Simple and Unified Baseline for Few-Shot Instance Segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2024;46(12):9221-9238. doi:10.1109/tpami.2024.3421340

BibTeX
@article{Han2024,
  author = {Han, Yue and Zhang, Jiangning and Wang, Yabiao and Wang, Chengjie and Liu, Yong and Qi, Lu and Yang, Ming-Hsuan and Li, Xiangtai},
  title = {Reference Twice: A Simple and Unified Baseline for Few-Shot Instance Segmentation},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2024},
  volume = {46},
  number = {12},
  pages = {9221--9238},
  doi = {10.1109/tpami.2024.3421340}
}
Kostenlos starten →