arXiv 2604.156782026-09-16cs.CV

HyCal：一种无需训练的跨学科少样本类增量学习原型校准方法

HyCal: A Training-Free Prototype Calibration Method for Cross-Discipline Few-Shot Class-Incremental Learning

Chung-Ang University（Chung-Ang 大学）；KT Corporation（KT 公司）；Dentium；ETRI

Eunju Lee, MiHyeon Kim, JuneHyoung Kwon, Yoonji Lee, JiHyun Kim, Soojin Jang, YoungBin Kim

备注：: Accepted to CVPR 2026. Eunju Lee and MiHyeon Kim contributed equally as co-first authors. Official code implementation is available at: https://github.com/EJLEE5826/HyCal-CIL

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出HyCal方法，通过结合余弦相似度和马氏距离，解决跨学科少样本类增量学习中的领域引力问题，提升在不平衡数据下的性能。

摘要：

预训练视觉-语言模型（VLMs）如CLIP在持续学习中表现出潜力，但现有少样本类增量学习（FSCIL）方法假设同质领域和平衡数据分布，限制了实际应用，因为数据来自异质学科，样本分布不平衡且视觉复杂性不同。我们识别出领域引力，即异质领域间的数据不平衡导致过代表或低熵领域 disproportionately 影响嵌入空间，导致原型漂移和在欠代表或高熵领域性能下降。为此，我们引入跨学科变量少样本类增量学习（XD-VSCIL）基准，捕捉真实世界的异质性和不平衡性，其中领域引力自然增强。我们提出混合原型校准（HyCal），一种无需训练的方法，结合余弦相似度和马氏距离，以捕捉互补的几何属性——方向对齐和协方差感知的幅度，从而在不平衡异质条件下获得稳定的原型。在冻结的CLIP嵌入上运行，HyCal实现了一致的保留-适应改进，同时保持高效。实验表明，HyCal有效缓解了领域引力，并在不平衡跨领域增量学习中优于现有方法。

英文摘要：

Pretrained Vision-Language Models (VLMs) like CLIP show promise in continual learning, but existing Few-Shot Class-Incremental Learning (FSCIL) methods assume homogeneous domains and balanced data distributions, limiting real-world applicability where data arises from heterogeneous disciplines with imbalanced sample availability and varying visual complexity. We identify Domain Gravity, a representational asymmetry where data imbalance across heterogeneous domains causes overrepresented or low-entropy domains to disproportionately influence the embedding space, leading to prototype drift and degraded performance on underrepresented or high-entropy domains. To address this, we introduce Cross-Discipline Variable Few-Shot Class-Incremental Learning (XD-VSCIL), a benchmark capturing real-world heterogeneity and imbalance where Domain Gravity naturally intensifies. We propose Hybrid Prototype Calibration (HyCal), a training-free method combining cosine similarity and Mahalanobis distance to capture complementary geometric properties-directional alignment and covariance-aware magnitude-yielding stable prototypes under imbalanced heterogeneous conditions. Operating on frozen CLIP embeddings, HyCal achieves consistent retention-adaptation improvements while maintaining efficiency. Experiments show HyCal effectively mitigates Domain Gravity and outperforms existing methods in imbalanced cross-domain incremental learning.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。