This repository contains the CRAG-MM dataset, a high-quality conversational benchmark for multimodal assistants. The dataset features conversations about images with varied complexity levels, designed to evaluate AI systems' visual understanding and conversational abilities.
CRAG-MM is a visual question-answering benchmark that focuses on factual questions, offering a unique collection of image and question-answering sets… See the full description on the dataset page:
https://huggingface.co/datasets/crag-mm-2025/crag-mm-single-turn-public.
