TsCA : on the semantic consistency alignment via conditional transport for compositional zero-shot learning

Abstract

Compositional Zero-Shot Learning (CZSL) aims to recognize novel state–object compositions by leveraging the shared knowledge of their primitive components. Despite considerable progress, effectively calibrating the bias between semantically similar multimodal representations, as well as generalizing pre-trained knowledge to novel compositional contexts, remains an enduring challenge. In this paper, we revisit conditional transport (CT) theory and its homology to visual–semantics interaction in CZSL, and propose a novel Trisets Consistency Alignment framework (TsCA) to address these issues. Concretely, we utilize three distinct yet semantically homologous sets—patches, primitives, and compositions—to construct pairwise CT costs that minimize semantic discrepancies. To further ensure consistency transfer within these sets, we implement a cycle-consistency constraint that refines learning by guaranteeing feature consistency of self-mapping during transport flow, regardless of modality. Moreover, we extend CT plans to an open-world setting, enabling the model to effectively filter out unfeasible pairs, thereby speeding up inference and improving accuracy. Extensive experiments verify the effectiveness of the proposed method. Code is available at https://github.com/keepgoingjkg/TsCA.