Researchers from UCLA and Stanford Introduce MRAG-Bench: An AI Benchmark Specifically Designed for Vision-Centric Evaluation for Retrieval-Augmented Multimodal Models