Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

FlowComposer: Composable Flows for Compositional Zero-Shot Learning

About

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. However, such PEFT-based designs suffer from two fundamental limitations: (1) Implicit Composition Construction, where composition is realized only via token concatenation or branch-wise prompt tuning rather than an explicit operation in the embedding space; (2) Remained Feature Entanglement, where imperfect disentanglement leaves attribute, object, and composition features mutually contaminated. Together, these issues limit the generalization ability of current CZSL models. In this paper, we are the first to systematically study flow matching for CZSL and introduce FlowComposer, a model-agnostic framework that learns two primitive flows to transport visual features toward attribute and object text embeddings, and a learnable Composer that explicitly fuses their velocity fields into a composition flow. To exploit the inevitable residual entanglement, we further devise a leakage-guided augmentation scheme that reuses leaked features as auxiliary signals. We thoroughly evaluate FlowComposer on three public CZSL benchmarks by integrating it as a plug-and-play component into various baselines, consistently achieving significant improvements.

Zhenqi He, Lin Li, Long Chen• 2026

Related benchmarks

TaskDatasetResultRank
Compositional Zero-Shot LearningC-GQA open world
HM Score12.6
65
Compositional Zero-Shot LearningUT-Zappos Closed World
HM58.6
57
Compositional Zero-Shot LearningC-GQA Closed World
HM34
56
Compositional Zero-Shot LearningUT-Zappos open world
HM51
52
Compositional Zero-Shot LearningMIT-States Closed World (test)
AUC23.5
27
Compositional Zero-Shot LearningMIT-States Open-world (test)
Seen Accuracy50.4
14
Showing 6 of 6 rows

Other info

Follow for update