Compositionality Benchmarks

该数据集是一个用于评估视觉和语言模型组合性的工具包，它融合了多种图像到文本以及文本到图像任务的基准测试。该工具包具有多样性任务，并旨在具备可扩展性，以便未来能够加入更多的基准测试和模型。它涵盖了不同规模的多项基准测试，其核心任务是进行组合性评估。

This dataset is a toolkit for evaluating the compositionality of vision-language models. It integrates benchmarks for multiple image-to-text and text-to-image tasks. Featuring a diverse set of tasks, this toolkit is engineered for scalability, allowing for the incorporation of additional benchmarks and models in future iterations. It covers multiple benchmarks of varying scales, with its core task being compositional evaluation.