AI RESEARCH
DeCLIP: Decoupled Prompting for CLIP-based Multi-Label Class-Incremental Learning
arXiv CS.CV
•
ArXi:2509.23335v2 Announce Type: replace Multi-label class-incremental learning (MLCIL) continuously expands the label space while recognizing multiple co-occurring classes, making it prone to catastrophic forgetting and high false-positive rates (FPR). Extending CLIP to MLCIL is non-trivial because co-occurring categories violate CLIP's single image-text alignment paradigm and task-level partial labeling induces high FPR. We propose DeCLIP, a replay-free and parameter-efficient framework that decouples CLIP representations via a one-to-one class-specific prompting scheme.