MM-BRIGHT is the first multimodal benchmark designed for reasoning-intensive retrieval. Unlike existing benchmarks that primarily consist of text-based, keyword-centric queries, MM-BRIGHT targets complex real-world scenarios where queries contain multimodal elements—such as diagrams, charts, and screenshots—that require deep reasoning to identify relevant documents.
Existing… See the full description on the dataset page:
https://huggingface.co/datasets/mm-bright/MM-BRIGHT.
