AI RESEARCH
FlowComposer: Composable Flows for Compositional Zero-Shot Learning
arXiv CS.CV
•
ArXi:2603.16641v1 Announce Type: new Compositional zero-shot learning (CZSL) aims to recognize unseen attribute-object compositions by recombining primitives learned from seen pairs. Recent CZSL methods built on vision-language models (VLMs) typically adopt parameter-efficient fine-tuning (PEFT). They apply visual disentanglers for decomposition and manipulate token-level prompts or prefixes to encode compositions.