CVLGJul 25

Breaking the Synthetic-Real Domain Shortcut for Training-Free Generative Replay-based Class Incremental Learning

arXiv:2607.229948.5Has CodeICML
Predicted impact top 48% in CV · last 90 daysOriginality Incremental advance
AI Analysis

For class-incremental learning practitioners, this work addresses a critical bottleneck in training-free generative replay, enabling privacy-preserving continual learning without performance degradation.

The paper identifies a 'domain shortcut' problem in generative replay for class-incremental learning, where models rely on domain differences rather than semantic features when mixing synthetic and real data. The proposed DREAM method eliminates this shortcut via subspace rectification and orthogonal projection, achieving state-of-the-art performance on 4 datasets.

Class-incremental learning (CIL) requires models to continuously acquire new knowledge while avoiding catastrophic forgetting. While exemplar replay is effective, it raises concerns regarding privacy and storage. Thus, generative replay has emerged as a viable alternative, synthesizing old data using frozen pretrained text-to-image (T2I) models without any extra training. However, we observe that directly mixing synthetic old-class data with real new-class data during incremental training leads to significant performance degradation. This issue stems from a "domain shortcut", where models rely on domain-discriminative features instead of semantic class cues. To address this, we propose DREAM ($\underline{\mathbf{D}}$omain-$\underline{\mathbf{R}}$egularized $\underline{\mathbf{E}}$xemplar-free $\underline{\mathbf{A}}$lignment $\underline{\mathbf{M}}$odel), which uses a training-free generator to synthesize old-class data and eliminates domain shortcut via subspace rectification and orthogonal projection, while reinforcing semantic alignment through real-anchored prototype regularization. Extensive experiments on 4 datasets demonstrate that DREAM outperforms existing exemplar-free CIL methods and achieves state-of-the-art performance. Our source code is available at https://github.com/Light-ZhangTao/DREAM.

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes