uclanlp/MRAG-Bench

MRAG-Bench: Vision-Centric Evaluation for Retrieval-Augmented Multimodal Models 🌐 Homepage | 📖 Paper | 💻 Evaluation Intro MRAG-Bench consists of 16,130 images and 1,353 human-annotated multiple-choice questions across 9 distinct scenarios, providing a robust and systematic evaluation of Large Vision Language Model (LVLM)’s vision-centric multimodal retrieval-augmented generation (RAG) abilities. Results Evaluated upon 10 open-source and 4… See the full description on the dataset page: https://huggingface.co/datasets/uclanlp/MRAG-Bench.