GENIUS: A Generative Framework for Universal Multimodal Search
This work addresses the need for efficient and versatile retrieval systems across multiple modalities and domains, representing a novel method for a known bottleneck rather than an incremental advance.
The paper tackles the problem of task-specific limitations and performance gaps in generative retrieval by proposing GENIUS, a universal generative retrieval framework that supports diverse multimodal tasks, achieving clear performance improvements over prior generative methods on the M-BEIR benchmark while maintaining high retrieval speed.
Generative retrieval is an emerging approach in information retrieval that generates identifiers (IDs) of target data based on a query, providing an efficient alternative to traditional embedding-based retrieval methods. However, existing models are task-specific and fall short of embedding-based retrieval in performance. This paper proposes GENIUS, a universal generative retrieval framework supporting diverse tasks across multiple modalities and domains. At its core, GENIUS introduces modality-decoupled semantic quantization, transforming multimodal data into discrete IDs encoding both modality and semantics. Moreover, to enhance generalization, we propose a query augmentation that interpolates between a query and its target, allowing GENIUS to adapt to varied query forms. Evaluated on the M-BEIR benchmark, it surpasses prior generative methods by a clear margin. Unlike embedding-based retrieval, GENIUS consistently maintains high retrieval speed across database size, with competitive performance across multiple benchmarks. With additional re-ranking, GENIUS often achieves results close to those of embedding-based methods while preserving efficiency.