PCogAlignBench

该数据集是一个包含18,000个实例和20个不同角色集的个体的基准测试，旨在评估视觉-语言模型在个性化对齐方面的表现。该基准测试根据角色所在位置被划分为两个子集：LS1和LS2，这有助于检验方法在训练和测试中的性能。数据集的规模为18,000个实例和20个个体，任务重点是基于情境认知评估视觉-语言模型的个性化对齐效果。

This dataset is a benchmark containing 18,000 instances and 20 individuals across distinct character sets, designed to evaluate the performance of Vision-Language Models (VLMs) on personalized alignment tasks. The benchmark is divided into two subsets, LS1 and LS2, based on the location of the characters, which helps examine the performance of methods during both training and testing phases. With a total scale of 18,000 instances and 20 individual subjects, the benchmark's core task focuses on evaluating the personalized alignment effectiveness of VLMs based on situational cognition.