Visual Document Retrieval Goes Multilingual

A new multilingual visual document retrieval system has been developed, enabling users to search and retrieve documents in various languages using visual queries. The system employs a transformer-based architecture with a shared multimodal embedding space, achieving state-of-the-art performance on multilingual benchmarks such as MMR and MMR-M. This advancement is significant for practitioners as it enhances the capabilities of AI systems in accessing and retrieving information across diverse languages and formats, thereby broadening the applicability of visual search technologies in global contexts.