Learning to Curate What You Generate for Generalizable Few-Shot Class-Incremental Learning

2026-10-04

AI Summary

This study addresses the challenges of representation degradation, semantic drift, and decision boundary instability in few-shot class-incremental learning under scarce base data by proposing the G-FSCIL framework. This method introduces a novel synthetic knowledge curation mechanism built upon a frozen latent diffusion model, which constructs a candidate pool via class inversion and distills a transferable selection strategy. Furthermore, a bidirectional boundary calibration scheme coupled with prototype initialization is designed to mitigate conflicts between novel and base classes. Experimental results demonstrate that the proposed framework significantly outperforms existing baselines, effectively suppressing catastrophic forgetting while enhancing the recognition balance between old and new categories.

Abstract

Few-shot class-incremental learning (FSCIL) aims to learn novel classes from limited annotations while preserving prior knowledge. Existing methods typically assume a sufficiently large base session, but this assumption fails when both base and incremental data are scarce, leading to weak initial representations, semantic drift, and unstable boundaries. We study this underexplored yet realistic setting, termed Generalizable FSCIL (G-FSCIL), where the base session itself contains only a few classes. Although synthetic data can alleviate supervision scarcity, naively mixing generated samples often introduces semantic noise and exacerbates old-new boundary conflicts. To address this, we propose a framework that curates trustworthy synthetic knowledge for stable G-FSCIL. Specifically, we first construct class-specific synthetic candidate pools using a frozen latent diffusion model, where class inversion is performed at the first observation and the resulting condition embeddings are reused for on-demand generation. Building on these candidates, we learn a knowledge curation strategy that selects samples with both semantic consistency and visual diversity, and distill this process into a transferable selection policy during the base session, which is then reused without further optimization. Leveraging the curated synthetic data, we further design a boundary-stable incremental adaptation scheme, including synthetic-informed prototype initialization and bidirectional boundary calibration to mitigate old-new conflicts. Extensive experiments demonstrate that our method consistently outperforms existing FSCIL baselines, with reduced forgetting and improved balance between old and new classes. Code is available at https://github.com/NiHaoWoJiaoYYC/G-FSCIL.

Problem

• Few-Shot Class-Incremental Learning
• Generalizable FSCIL
• Synthetic Data Curation
• Semantic Drift
• Boundary Conflict

Innovation

• Generalizable Few-Shot Class-Incremental Learning
• Latent Diffusion Model
• Knowledge Curation
• Boundary Calibration
• Synthetic Data