CVPR Findings2026
Towards Complete Activation: Foreground-Background Multi-Perspective Guided Cross-Support for Few-Shot Segmentation
Yi Yang, Qiang Jiao, Mengrui Shi, Qiang Zhang
Abstract
Few-shot semantic segmentation (FSS) aims to segment novel classes based on only a few annotated samples. Despite recent advances, existing methods still suffer from incomplete activation, as they often focus on the most salient regions while overlooking semantically relevant but less conspicuous parts. To address this issue, we propose a Foreground-Background Multi-Perspective Guided Cross-Support Network (FB-CSNet) that improves activation completeness through a novel locate-then-cross-support strategy. Specifically, we first introduce a Multi-Prompt Foreground-Background Localization (MFBL) module that leverages diverse descriptive text prompts to accurately localize both foreground and background regions in complex scenes. Building on these localizations, we further propose a Multi-Perspective Cross-Support (MPCS) module. This module uses an Adaptive Query Transformation (AQT) mechanism to generate diverse query variants with enhanced foreground emphasis. These transformed views, together with the original query image, interact via a bidirectional cross-support mechanism to mutually reinforce activations across perspectives, leading to more complete foreground coverage. Extensive experiments on standard FSS benchmarks demonstrate the effectiveness of our approach, with 80.8% mIoU on the PASCAL-5 i dataset and 61.0% mIoU on the COCO-20 i dataset under the 1-shot setting, establishing new state-of-the-art performance.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- Image Segmentation Using Text and Image Prompts cites Language Models are Few-Shot Learners
- Image Segmentation Using Text and Image Prompts cites Learning Transferable Visual Models From Natural Language Supervision
- Learning Transferable Visual Models From Natural Language Supervision cites Language Models are Few-Shot Learners
Builds on14
- Language Models are Few-Shot LearnersTom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah et al.NeurIPS 2020 · 64,255 citations
- Learning Transferable Visual Models From Natural Language SupervisionAlec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh et al.ICML 2021 · 47,906 citations
- Expectation-Maximization Attention Networks for Semantic SegmentationXia Li, Zhisheng Zhong, Jianlong Wu, Yibo Yang et al.ICCV 2019 · 639 citations
- Image Segmentation Using Text and Image PromptsTimo Lüddecke, Alexander S. EckerCVPR 2022 · 457 citations
- Feature Weighting and Boosting for Few-Shot SegmentationKhoi Nguyen, Sinisa TodorovicICCV 2019 · 402 citations
Related papers
- Focus on Query: Adversarial Mining Transformer for Few-Shot SegmentationYuan Wang, Naisong Luo, Tianzhu ZhangNeurIPS 2023 · 29 citations
- Object-Level Correlation for Few-Shot SegmentationChunlin Wen, Yu Zhang, Jie Fan, Hongyuan Zhu et al.ICCV 2025 · 5 citations
- Integrative Few-Shot Learning for Classification and SegmentationDahyun Kang, Minsu ChoCVPR 2022 · 76 citations
- Adaptive FSS: A Novel Few-Shot Segmentation Framework via Prototype EnhancementJing Wang, Jiangyun Li, Chen Chen, Yisi Zhang et al.AAAI 2024 · 24 citations
- CGMGM: A Cross-Gaussian Mixture Generative Model for Few-Shot Semantic SegmentationJunao Shen, Kun Kuang, Jiaheng Wang, Xinyu Wang et al.AAAI 2024 · 10 citations
