MegaPairs: Massive Data Synthesis For Universal Multimodal Retrieval
This addresses a data bottleneck for multimodal retrieval researchers, offering a scalable solution with public resources, though it is incremental in leveraging existing VLMs and images.
The paper tackles the lack of training data in multimodal retrieval by introducing MegaPairs, a data synthesis method that generates over 26 million instances, enabling models to outperform baselines trained on 70× more data and achieve state-of-the-art zero-shot performance on 4 CIR benchmarks and MMEB.
Despite the rapidly growing demand for multimodal retrieval, progress in this field remains severely constrained by a lack of training data. In this paper, we introduce MegaPairs, a novel data synthesis method that leverages vision language models (VLMs) and open-domain images, together with a massive synthetic dataset generated from this method. Our empirical analysis shows that MegaPairs generates high-quality data, enabling the multimodal retriever to significantly outperform the baseline model trained on 70$\times$ more data from existing datasets. Moreover, since MegaPairs solely relies on general image corpora and open-source VLMs, it can be easily scaled up, enabling continuous improvements in retrieval performance. In this stage, we produced more than 26 million training instances and trained several models of varying sizes using this data. These new models achieve state-of-the-art zero-shot performance across 4 popular composed image retrieval (CIR) benchmarks and the highest overall performance on the 36 datasets provided by MMEB. They also demonstrate notable performance improvements with additional downstream fine-tuning. Our produced dataset, well-trained models, and data synthesis pipeline will be made publicly available to facilitate the future development of this field.