MIRAGE
收藏资源简介:
MIRAGE是一个多模态信息寻求和推理基准,专为农业领域设计。该数据集由超过35,000个真实用户与专家的交互组成,涵盖了作物健康、病虫害诊断和作物管理等场景。数据集包括超过7,000个独特的生物实体,包括植物种类、害虫和疾病,是视觉语言模型领域中最具分类多样性的基准之一。MIRAGE的特点是未指定、上下文丰富的场景,要求模型推断潜在的知识差距,处理稀有实体,并主动引导交互或做出响应。该数据集旨在解决现实世界中知识密集型领域的专家咨询问题。
MIRAGE is a multimodal information-seeking and reasoning benchmark tailored for the agricultural domain. This dataset comprises over 35,000 real-world user-expert interactions, encompassing scenarios including crop health, pest and disease diagnosis, and crop management. It contains over 7,000 distinct biological entities, including plant species, pests, and diseases, ranking among the most categorically diverse benchmarks in the vision-language model field. MIRAGE features unspecified, context-rich scenarios that mandate models to infer latent knowledge gaps, handle rare entities, and proactively initiate interactions or generate appropriate responses. This benchmark aims to address expert consultation challenges in knowledge-intensive real-world domains.

- 1MIRAGE: A Benchmark for Multimodal Information-Seeking and Reasoning in Agricultural Expert-Guided Conversations伊利诺伊大学厄巴纳-香槟分校、亚马逊 · 2025年



