VisualSphinx/VisualSphinx-V1-Raw

VisualSphinx是一个大规模的合成视觉逻辑谜题开源数据集，包含超过660K个自动生成的逻辑视觉谜题。每个逻辑谜题都配有一个可解释的规则，并伴有正确答案和合理的干扰项。数据集分为三个部分：110k_4options（包含超过110K个具有4个选项的视觉逻辑谜题）、440k_4options_shuffle（包含超过440K个打乱答案顺序的视觉逻辑谜题）和110k_10options（包含超过110K个具有10个选项的视觉逻辑谜题）。每个谜题包括唯一的标识符、视觉谜题图像、标准化的问题提示、多个选择选项、正确答案、详细的规则解释、是否有重复图像的标志、逻辑连贯性评分、可读性评分和通过率。

VisualSphinx is a large-scale synthetic dataset of vision logic puzzles, containing over 660K automatically generated logical visual puzzles. Each logical puzzle is grounded with an interpretable rule and accompanied by both correct answers and plausible distractors. The dataset is split into three parts: 110k_4options (over 110K visual logic puzzles with a 4-option multiple choice format), 440k_4options_shuffle (over 440K visual logic puzzles with a shuffled answer format), and 110k_10options (over 110K visual logic puzzles with a 10-option multiple choice format). Each puzzle includes a unique identifier, visual puzzle images, standardized puzzle prompts, multiple choice options, correct answers, detailed rule-based explanations, a flag indicating duplicate images, a logical coherence score, a readability score, and a pass rate.