ScienceVQA
收藏ScienceVQA数据集概述
数据集基本信息
- 名称: ScienceVQA
- 语言: 英语 (en)
- 许可证: Apache 2.0
- 标签: Science, VQA, Image Quality
- 规模: 1K<n<10K
- 数据来源: 原始数据集 (original)
数据集结构与配置
数据集包含三个配置。
配置一: SIQA-S
- 用途: 科学图像质量评分
- 特征:
pid: 整型标识符image: 图像数据image_path: 图像路径字符串perception_raing: 感知评分 (浮点型)knowledge_rating: 知识评分 (浮点型)
- 数据划分:
train: 8,400 个样本,占用空间约 1.49 GB
- 下载大小: 约 1.71 GB
- 数据集大小: 约 1.49 GB
配置二: SIQA-U
- 用途: 科学视觉问答 (VQA)
- 特征:
image: 图像数据image_path: 图像路径字符串class: 类别字符串category: 类别字符串type: 问题类型字符串question: 问题字符串option: 选项字符串answer: 答案字符串explantion: 解释字符串difficulty: 难度字符串is_wrong: 布尔值precision: 精度字符串
- 数据划分:
train: 104,021 个样本,占用空间约 25.07 GB
- 下载大小: 约 22.52 GB
- 数据集大小: 约 25.07 GB
配置三: default
- 特征:
file_name: 图像数据
数据实例示例
评分数据实例 (SIQA-S)
json { "pid": 1, "image": "images/Mathematical Representation/6637_AnnualCrop_2941.jpg", "image_path": "images/Mathematical Representation/6637_AnnualCrop_2941.jpg", "perception_raing": 1.0, "knowledge_rating": 1.0 }
问答数据实例 (SIQA-U)
json { "image": "images/Data Visualization/3161_v4_v5_f67bce80a1a40d81c93fbfaac4df57399a86a3e95f135d83bbbe62d4e8a51553.jpg", "image_path": "images/Data Visualization/3161_v4_v5_f67bce80a1a40d81c93fbfaac4df57399a86a3e95f135d83bbbe62d4e8a51553.jpg", "class": "Data Visualization", "category": "completeness and conciseness", "type": "yes-or-no", "question": "Does the image include units for the axes?", "option": "A. Yes B. No", "answer": "B", "explantion": "The description of bad aspects states the image lacks units for both axes.", "difficulty": "easy", "is_wrong": false }




