ryokamoi/VisOnlyQA_Train

VisOnlyQA数据集旨在评估大型视觉语言模型（LVLMs）在科学图表几何信息上的视觉感知能力。该数据集包含12个视觉感知任务中的1200个多项选择题，涵盖了4类科学图表。此外，数据集还提供了包含7万个实例的训练数据。数据集分为评估集和训练集，评估集包括真实数据和合成数据，训练集则全部为合成数据。每个数据实例包含图像、问题、提示、答案等特征，并提供了详细的元数据信息。

VisOnlyQA is designed to evaluate the visual perception capability of large vision language models (LVLMs) on geometric information of scientific figures. The evaluation set includes 1,200 multiple choice questions in 12 visual perception tasks on 4 categories of scientific figures. Additionally, a training dataset consisting of 70k instances is provided. The dataset is divided into evaluation and training sets, with the evaluation set including both real and synthetic data, and the training set consisting entirely of synthetic data. Each data instance includes features such as images, questions, prompts, answers, and detailed metadata.