VisualSphinx/VisualSphinx-V1-Benchmark

VisualSphinx是一个包含660K个自动生成的逻辑视觉谜题的大规模合成开源数据集，旨在作为视觉逻辑推理能力的官方基准和评估套件。每个谜题都有可解释的规则和正确答案，并伴有合理的干扰项。数据集分为测试集，包含全面的注释，用于基准测试。

VisualSphinx is a large-scale synthetic open-source dataset containing 660K automatically generated logical visual puzzles, designed to serve as the official benchmark and evaluation suite for visual logic reasoning capabilities. Each puzzle is grounded with an interpretable rule and accompanied by correct answers and plausible distractors. The dataset includes a test split with comprehensive annotations for benchmarking.