UniParser/SpecVQA
收藏Hugging Face2025-12-26 更新2026-01-03 收录
下载链接：
https://hf-mirror.com/datasets/UniParser/SpecVQA
下载链接
链接失效反馈官方服务：
资源简介：
SpecVQA是一个专注于科学图像中光谱理解和视觉问答的基准数据集。该数据集旨在解决多模态大语言模型（MLLMs）在光谱解释中的关键失败模式。通过聚焦七种基本光谱类型，它提供了对科学准确性和领域知识使用的简洁而严格的评估。数据集包含由领域专家手动筛选的620张光谱图，每张图配有5个问题-答案对，共计3,100个QA对。这些问题分为描述性问题和推理性问题，支持中文和英文两种语言。
SpecVQA is a benchmark for spectral understanding and visual question answering in scientific images. It addresses the key failure modes of Multimodal Large Language Models (MLLMs) in spectral interpretation by focusing on seven essential spectrum types. The dataset includes 620 spectral figures curated by domain experts, each accompanied by 5 question-answer pairs, totaling 3,100 QA pairs. These questions are categorized into descriptive and reasoning questions, and the dataset supports both Chinese and English languages.
提供机构：
UniParser
