#omni-modal retrieval

Conan-embedding-v3: Fusing Modality-Specific Models for Omni-Modal Embedding

Shiyu Li, Zhiyuan Hu, Yifan Wang +3

The paper presents Conan-embedding-v3, a framework that trains modality‑specific specialist models, fuses them into a single dense backbone, and then recovers the modality projecto…