ARO

该数据集包含了一系列测试，旨在衡量模型在编码物体与属性之间的组合关系方面的能力。此外，该数据集用于评估稳定扩散模型的表现，涉及将真实标题与经过干扰的版本进行比较，以完成对模型组合理解能力的评估任务。

This dataset includes a series of tests intended to measure a model's ability to encode compositional relationships between objects and their attributes. Furthermore, this dataset is used to evaluate the performance of Stable Diffusion models, where the evaluation task involves comparing ground-truth captions with their perturbed versions to assess the model's compositional understanding capabilities.